Incorporating Splines and Offset in MixedModels.jl

Specific Domains Statistics
1

Hi,

I am working on project that required doing multilevel models and the dataset is quite large (7 million rows). My current implementation is in R but takes quite a bit of time using glmmTMB package. So, I was wondering whether this can be replicated using MixedModels.jl with the hopes that it will be faster.

My main issue is how to add an offset term and a spline. As for the offset I saw some old posts stating that it hasn’t been implemented but wondering if it has by now.

Here is my current R implementation

library(glmmTMB)
glmm_formula_hosp <- hospitalisation_cum_lead7 ~  
  publicholiday +
  dow +
  splines::ns(date) +
  hw_mean_ehf_severity +
  age_band3 +
  sex +
  (1 | sa2_2021_code) +
  stats::offset(log(population_interp))

m_glmm <- glmmTMB::glmmTMB(
  formula = glmm_formula_hosp,
  data = df,
  family = glmmTMB::nbinom1(link = "log")
)
2

Post from four years ago discusses offsets.. Offsets are available. Where you will have to make a decision is whether to use Poisson() as an approximation to the negative binomial if overdispersion is mild. Splines2 provides an equivalent to R’s ns. If negative binomial is a must-have, there’s Turing but performance will be slow.