countreg

Control Parameters for Hurdle Count Data Regression

Description

Various parameters that control fitting of hurdle regression models using hurdle.

Usage

hurdle.control(method = "BFGS", maxit = 10000, trace = FALSE,
  separate = TRUE, start = NULL, hessian = TRUE, ...)

Arguments

`method`	characters string specifying the `method` argument passed to `optim`.
`maxit`	integer specifying the `maxit` argument (maximal number of iterations) passed to `optim`.
`trace`	logical or integer controlling whether tracing information on the progress of the optimization should be produced (passed to `optim`).
`separate`	logical. Should the estimation of the parameters in the truncated count component and hurdle zero component be carried out separately? See details.
`start`	an optional list with elements `“count”` and `“zero”` (and potentially `“theta”`) containing the coefficients for the corresponding component.
`hessian`	logical. Should the Hessian be computed to derive an estimate of the variance-covariance matrix? If `FALSE`, the variance-covariance matrix contains only `NA`s.
`…`	arguments passed to `optim`.

Details

All parameters in hurdle are estimated by maximum likelihood using optim with control options set in hurdle.control. Most arguments are passed on directly to optim, only trace is also used within hurdle and separate/start control how optim is called.

Starting values can be supplied via start or estimated by glm.fit (default).

If separate = TRUE (default) the likelihoods of the truncated count component and the hurdle zero component will be maximized separately, otherwise the joint likelihood is set up and maximized. In case of separate = FALSE and both dist == “negbin” and zero.dist == “negbin” the theta parameter is restricted to be identical across both negative binomial distributions.

Standard errors are derived numerically using the Hessian matrix returned by optim. To supply starting values, start should be a list with elements “count” and “zero” and potentially “theta” (a named vector, for models with negative binomial components only) containing the starting values for the coefficients of the corresponding component of the model.

Value

A list with the arguments specified.

Examples

library("countreg")

data("CrabSatellites", package = "countreg")

## default start values
fm1 <- hurdle(satellites ~ width + as.numeric(color), data = CrabSatellites,
  dist = "negbin", zero = "negbin")

## user-supplied start values and other options
fm2 <- hurdle(satellites ~ width + as.numeric(color), data = CrabSatellites,
  dist = "negbin",
  zero = "negbin",
  trace = TRUE,
  separate = FALSE,
  start = list(count = c(0.5, 0, 0),
           zero = c(-10, 0.5, -0.5),
           theta = c(count = 1, zero = 1)))

Hurdle Count Model
count model: negbin with log link
zero hurdle model: negbin with log link
dependent variable:
 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 
62 16  9 19 19 15 13  4  6  3  3  1  1  0  1  1 
calling optim() for joint count and zero hurlde estimation:
initial  value 408.872181 
iter  10 value 350.643112
final  value 350.367259 
converged
done