Skip to contents

Generic random intercepts (1 | group) (bar syntax)

Source: R/re-int.R
re_int.Rd

A gllvmTMB() formula token, modelled on lme4 / glmmTMB syntax, that adds a per-row random intercept indexed by group. For each row o, the linear predictor gains u[group(o)], where u is a vector of length nlevels(group) with i.i.d. prior u_g ~ N(0, sigma^2). Multiple bar terms are allowed, each with its own variance component.

Arguments

condition

Bar expression of the form 1 | group. The right-hand side group must be the unquoted name of a column in data; if the column is not already a factor it is coerced to one.

Value

A formula marker; never evaluated as a call. The token is recognised by gllvmTMB()'s formula parser via the parens that wrap 1 | group.

Details

What is and isn't supported

Only the random-intercept form (1 | group) is implemented in this release. The natural extensions are not yet available:

  • (0 + x | group) — random slopes (no intercept). Coming in a future release.

  • (1 + x | group) — correlated intercept + slope. Coming in a future release.

  • (0 + trait | group) — trait-specific random intercepts. Use the existing unique() / latent() covstructs for that pattern.

If you write any of these unsupported forms, gllvmTMB() aborts with a message naming the unsupported form rather than silently fitting the wrong model.

Usage

Inside a gllvmTMB() formula, alongside any of the existing covstruct terms:

# Add a study-level random intercept on top of a latent() block:
fit <- gllvmTMB(value ~ 0 + trait + latent(0 + trait | site, d = 2) +
                         (1 | study),
                data = df)

# Two random intercepts (e.g. study + dataset):
fit <- gllvmTMB(value ~ 0 + trait + (1 | study) + (1 | dataset),
                data = df)

The fitted variance components live at fit$report$log_sigma_re_int (one entry per (1|...) term, in the order the terms appear in the formula); take exp() to get sigma. The fitted random-intercept BLUPs live at fit$tmb_obj$env$parList()$u_re_int, packed by term in the same order; offsets and per-term lengths are recorded in fit$re_int.