Censored data reasonable prediction


My response looks like this and in essence is a instrument that cannot measure above a maximum. On the other hand we know that the response cannot go too too far away from the top limit.

If I plot the response against the predictors look like this,

I’ve been told that would be reasonable to handle it as censored data so I rescaled the response to [0, 1] and I run something looking like that bf(y1|cens(y2) ~ var1 + var2)

However, I don’t like what I see from the conditional_effects plot because I would expect to get a plateau at a point and should be no results above 1.4.

My questions is how can let the model know that the response has a threshold in 1.4 units?

Thank you

This may or may not be appropriate for your case but one potential solution would be to model the deviation from the maximum. If y is your current outcome, you could model ymax - y. It might be that you need a zero-inflated and/or a lognormal distribution to fully capture all the features of the underlying process. The advantage is that this approach might be more flexible than the censored model.