Per-cluster score matrix

The gradient of each cluster's contribution to the marginal log-likelihood, evaluated at the fitted estimates. This is the quantity sandwich::estfun() would return if the marginalized objective had per-observation contributions, which it does not - see vcov_cluster() for the argument and the guards.

Usage

cluster_scores(object, cluster)

Arguments

object

A frmtmb_fit.

cluster

The clustering factor: a one-sided formula such as ~ g (looked up in the model frame, then in the data the model was fitted to), a variable name, or a vector with one entry per fitted row.

Value

A matrix with one row per cluster level and one column per estimated outer parameter, named as in vcov(object, full = TRUE). Signs follow sandwich::estfun(): these are scores of the log-likelihood, not of the objective.

Details

Rows sum to the gradient of the full objective, which is (near) zero at a converged optimum; that identity is the cheapest check that the clustering factor really does split the likelihood.

See also

vcov_cluster(), which sandwiches these between two copies of the inverse observed information.

Examples

set.seed(1)
dd <- data.frame(x = rnorm(200), g = factor(rep(1:25, 8)))
dd$y <- rnorm(200, 1 + 0.5 * dd$x + rnorm(25, 0, 0.8)[dd$g], 1)
fit <- frm(bf(y ~ x + (1 | g)) + gaussian(), data = dd, REML = FALSE)

S <- cluster_scores(fit, ~ g)
dim(S)
#> [1] 25  4
# the scores add up to the gradient at the optimum
max(abs(colSums(S)))
#> [1] 2.344908e-05