View source: R/FUN_glmm_families.R
| nbinom2 | R Documentation |
Creates a sommer-native negative-binomial family with variance
\mu+\mu^2/\theta. It can be used in native Laplace fits without another
package providing the family or fitting engine.
nbinom2(theta = 1, link = "log")
theta |
Positive starting shape. Larger values approach a Poisson distribution. |
link |
Currently |
For mmes() with method$integration="laplace", shape is estimated
jointly with fixed effects and free random covariance coordinates by default.
Use familyParameters=list(theta=2, estimate=FALSE) to fix shape.
For familym(), nest this configuration under the trait's name.
The constructor's theta is the default starting value; an explicitly
supplied control value overrides it. Counts must be non-negative integers.
The full likelihood includes the negative-binomial normalization constants.
Native derivatives include shape dependence of both the conditional mode and
Laplace determinant correction. The fitted shape and approximate uncertainty
are available as theta.nb, theta.nb.se, and
familyParameters for a single-family fit. Uncertainty requires a
converged, interior fit with valid joint curvature. Shape near the Poisson
limit can be weakly identified; numerical boundary/curvature warnings are not
calibrated hypothesis tests.
This family also supplies the ordinary R family methods needed by PQL, where shape is fixed unless the separate PQL shape-estimation control is enabled. PQL is not changed into a marginal-likelihood estimator by choosing this family.
An object of class "family".
mmes, familym, mmes_control
family <- nbinom2(theta=2)
family$variance(c(1, 2, 3))
method <- mmes_method(objective="ml", integration="laplace")
# Pass familyParameters=list(theta=2, estimate=FALSE) to mmes().
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.