Enforce the maintenance on a given state

[NunoSH]

Hi Brett,

quick question: is there any way to enforce the maintenance on the same state?
For example, I have 5 states, and I want to enforce the maintenance on state 3 and 4 once each of these states start and have no say on the remaining transitions. So in essence, I think, what I want is to increase the probability on the non-transition in state 3 and 4 (or in the transition 3->3 and 4->4).

As I have seen, we can influence the transitions from and to different states using "beta" on the Fixpar, or formula, but I am not sure if or how I could do the opposite.

I hope I made myself clear :)

Thank you very much!

Nuno

[bmcclintock]

I'm not sure I understand what you're wanting to do, but if you're trying to make both states 3 and 4 absorbing (i.e., once an individual enters state 3 or 4, then no more state switches can occur), then you can do this using the fixPar argument in fitHMM. For example,

fixPar <- list(beta=matrix(c(NA,NA,NA,NA,
                             NA,NA,NA,NA,
                             -1.e+100,-1.e+100,-1.e+100,-1.e+100,
                             -1.e+100,-1.e+100,-1.e+100,-1.e+100,
                             NA,NA,NA,NA),1,5*4))

would effectively prevent switches out of states 3 and 4.

[NunoSH]

Good morning Brett,

thanks a lot for your reply!

In a sense, it is more or less what you suggested, but with the possibility of switching out from these states. I would like to make it "harder" to switch out from these states but not impossible.

Because in order to an individual to get into state 3 or 4, there needs to be the presence (1-presence, 0-absence) of a feature (modeled with the Bernoulli distribution) and the probabilities of presence of these features (in state 3 and 4) have a really high values (>0.9). But what sometimes happens is that there is a punctual absence of these features, but I know the individuals are still in one of those states, but the HMM result gives that there is a switch out of one of these states, when actually there isn't. So I would like that these punctual absences of these features hadn't as much influence on switch off from state 3 or 4.

So to tackle this issue, is there a way of doing it in the MomentuHMM, or you thing it is better to address the data itself? (like doing a sort of smoothing or moving average for binary variables?

Thanks again!
I wish you a great day!

Nuno

[bmcclintock]

It's still hard for me to figure out what you're trying to do without much more detail. If you can provide a complete description of the data, model, and what you're trying to accomplish then I might be able to provide some suggestions. Code would also help. Based on your description, it sounds like you somehow know the individuals are still in state 3 or 4 even when there is a "punctual absence of these features". I don't know what "punctual absence of these features" means, but, assuming this means they are either missing (i.e. NA) or 0, what additional information do you have that indicates they must still be in state 3 or 4? This additional information would probably need to be incorporated into the model, either through additional (possibly constructed) covariates on the state transition probabilities or additional data streams.