Per-group coefficients (fixed effects plus conditional modes)

Follows the lme4/glmmTMB/brms convention: for each random-effect grouping factor, the fixed effects of its linear predictor broadcast over the group levels, with the conditional modes added to the matching columns. Use fixef() for the fixed effects alone.

Usage

# S3 method for class 'frmtmb_fit'
coef(object, summary = TRUE, robust = FALSE, probs = c(0.025, 0.975), ...)

Arguments

object: A frmtmb_fit.
summary, robust, probs: brms's arguments, in brms's positions so that a positional brms call asks the same question. brms answers summary = FALSE with the posterior draws and robust = TRUE with their median and MAD, and a maximum-likelihood fit has no draws, so both are refused by name with the reason. The default of each is accepted and changes nothing.
...: Refused: an argument the method does not have is an error naming it, rather than silently changing nothing.

Value

A named list of data frames, one per grouping factor, each with one row per group level and one column per coefficient. When random effects appear in more than one linear predictor, the list is nested one level deeper, keyed as in fixef(). A fit without random effects returns the per-predictor coefficient vectors instead.

Details

The result is a list of data frames keyed by grouping factor. When random effects appear in more than one dpar (or response), an outer layer keyed like fixef() is added. Smooth terms are excluded. A fit without random effects returns the coefficient vector of the location predictor (when there is one linear predictor), or one vector per predictor.

The coefficients are named as brms names them, which is how fixef() and vcov() name their rows: Intercept, not (Intercept). Anything that pairs coef() with vcov() by name, lmtest::coeftest() among them, needs the two to agree. Use fixef_by_dpar() for the design-column spelling.

Examples

set.seed(1)
dd <- data.frame(x = rnorm(100), g = factor(rep(1:10, 10)))
dd$y <- rnorm(100, 1 + 0.5 * dd$x + rnorm(10, 0, 0.8)[dd$g], 1)
fit <- frm(bf(y ~ x + (1 | g)) + gaussian(), data = dd)

# one row per group: the fixed effects with the modes added in
head(coef(fit)$g)
#>     Intercept        x
#> 1  0.44664608 0.670691
#> 2  1.43337185 0.670691
#> 3  0.54267978 0.670691
#> 4  0.87303616 0.670691
#> 5 -0.02537336 0.670691
#> 6  2.28544337 0.670691
# which is fixef() plus ranef(), the lme4 identity, under the same
# names fixef() and vcov() use
all.equal(coef(fit)$g[["Intercept"]],
          fixef(fit)["Intercept", "Estimate"] + ranef(fit)$g[, 1],
          check.attributes = FALSE)
#> [1] TRUE

# without random effects there are no groups, so coef() is the
# coefficient vector of the location predictor
coef(frm(bf(y ~ x) + gaussian(), data = dd))
#> Intercept         x 
#> 1.2457711 0.6136202