[Help] Hierarchical logistic regression with time-varying coefficients

Hi, has anyone seen any Stan example of hierarchical logistic regression with time-varying coefficients, captured by a dynamic linear model (Kalman filter type) process? Thanks.

Does this help Example von Bertalanffy model (and hierarchical logistic regression and linear regression) as a starting point?

1 Like

Check out Eric Ward’s course Applied Time Series Analysis (atsa-es.github.io)

He has moved much of the R MARSS package to Stan models that are part of the course he co-teaches (the above link goes to those packages as well).

Edit: I think this model example might be what you are looking for: atsar | Applied time series analysis in R with Stan. Allows fast Bayesian fitting of multivariate time-series models. (atsa-es.github.io)

2 Likes

Thanks for the replies thus far. Let me try to write down the model I have in mind:

Y[i, t]: binary outcome (1/0) for period i at time t
Y[i, t] = P( A[i, t] )
P(.) = inverse-logit
A[i, t+1] = A[i, t] + e[i, t+1]
e[i, t] ~ N ( 0, s ), for all t
A[i, 0] ~ N ( pop_A, pop_s )

I call this a hierarchical logistic time-varying random intercept model. Thanks.

I’m not sure what you are trying to do. But, if you want to do a latent, binary variable, your model seems similar to an occupancy model.

Checkout this package I have written as well: UMESC / quant-ecology / occStan · GitLab (usgs.gov)

As a tip, I suggest these steps:

  1. Writing out the math of your model
  2. Simulating your data
  3. Build up simple parts of your Stan model

Thanks. Could you tell me whether this is the stan code you are referring to? I use cmdStan so I cannot run your RStan-linked package. Thank you.

You found one model. There is also a two-level model in the same folder.