Skip to content

Structural dependence overview

Status — Stable

Mirrors drmTMB's Structural dependence overview. In DRM.jl today: all four structured-effect markers on the mean — relmat (supplied K), animal (pedigree A), phylo (tree), and spatial (coordinates) — are available. For Gaussian responses with these effects on the mean, latent effects integrate out exactly; covariance parameters are estimated numerically. The q=4 phylogenetic bivariate engine is also available.

A structured effect puts a known correlation among the group-level random intercepts instead of treating groups as independent. The marker names where that correlation comes from:

MarkerCorrelation sourceSupply
relmat(1 | id)an arbitrary known matrix KK = …
animal(1 | id)an additive-genetic pedigreeA = …
phylo(1 | species)a phylogenetic treetree = …
spatial(1 | site)distance between coordinatescoords = … (range estimated)

For Gaussian responses with Gaussian structured effects entering the mean linearly and residual variance independent of those effects, all four have the exact marginal covariance V = D + σ_s² Z K Zᵀ. No Laplace approximation is needed in this case. Covariance parameters are still estimated numerically. Each has its own worked tutorial — see below.

See also