Mixed effects model accounting for spatial lag: GPS points are at the second level

Modeling
,
1

Hi, I’m new here and absolutely love the brms package.

I wish to account for spatial autocorrelation in a model I’m specifying, but there’s a hitch: many observations share the same GPS location, which is at the same level as a grouping variable. There are a couple scenarios for this 1) using survey data where I only have the GPS location of a cluster point, and 2) health facility data where I have client-level (dichotomous) outcomes predicted based on health facility characteristics, only have gps locations for health facilities.

I’m curious if this is possible to implement in brms, e.g. if it’s possible to use sar() at the second level. In its simplest, non-spatial form, the code is currently:

brm(
    individualOutcome ~ individualPredictor + healthFacilityPredictor + (1|healthFacility_ID),
    data = dataset,
    family = bernoulli(link = "logit"),
    warmup = 6e3, 
    iter = 1.2e4, 
    seed = 123, 
    cores = 8,
    threads = threading(2),
    backend = "cmdstanr",
    chains = 8
  )

Additionally, how would marginal_effects handle this?

Huge thanks!