Native negative-binomial GLM (reproduces MASS::glm.nb)

Fits a negative-binomial GLM by the same alternating scheme as MASS::glm.nb: an initial Poisson fit, then iterate between a glm.fit with the current negative.binomial(theta) family and a theta MLE update until the log-likelihood and theta converge. Coefficients and theta match MASS::glm.nb to convergence tolerance.

Usage

morie_glm_nb(
  formula,
  data,
  weights,
  init.theta = NULL,
  link = "log",
  control = stats::glm.control(...),
  ...
)

Arguments

formula, data: Model formula and data.
weights: Optional prior weights.
init.theta: Optional starting theta.
link: Link (default "log").
control: A stats::glm.control object.
...: Passed to glm.control.

Value

A glm/negbin object with $theta.

References

Venables, W. N., & Ripley, B. D. (2002). Modern Applied Statistics with S. Springer.

Examples

set.seed(1); n <- 300
x <- rnorm(n)
y <- rnbinom(n, mu = exp(0.3 + 0.9 * x), size = 3)
fit <- suppressWarnings(morie_glm_nb(y ~ x, data = data.frame(y, x)))
coef(fit)
#> (Intercept)           x 
#>   0.3398887   0.7982631