Adaptive likelihood profiling.

Calculate 1D likelihood profiles wrt. single parameters or more generally, wrt. arbitrary linear combinations of parameters (e.g. contrasts).

tmbprofile(
  obj,
  name,
  lincomb,
  h = 1e-04,
  ytol = 2,
  ystep = 0.1,
  maxit = ceiling(5 * ytol/ystep),
  parm.range = c(-Inf, Inf),
  slice = FALSE,
  adaptive = TRUE,
  trace = TRUE,
  ...
)

Arguments

obj

Object from MakeADFun that has been optimized.

name

Name or index of a parameter to profile.

lincomb

Optional linear combination of parameters to profile. By default a unit vector corresponding to name.

h

Initial adaptive stepsize on parameter axis.

ytol

Adjusts the range of the likelihood values.

ystep

Adjusts the resolution of the likelihood profile.

maxit

Max number of iterations for adaptive algorithm.

parm.range

Valid parameter range.

slice

Do slicing rather than profiling?

adaptive

Logical; Use adaptive step size?

trace

Trace progress? (TRUE, or a numeric value of 1, gives basic tracing: numeric values > 1 give more information)

...

Unused

Value

data.frame with parameter and function values.

Details

Given a linear combination $$ t = \sum_{i=1}^n v_i \theta_i $$ of the parameter vector \(\theta\), this function calculates the likelihood profile of \(t\). By default \(v\) is a unit vector determined from name. Alternatively the linear combination may be given directly (lincomb).

See also

plot.tmbprofile, confint.tmbprofile

Examples

if (FALSE) { # \dontrun{
runExample("simple",thisR=TRUE)
## Parameter names for this model:
## beta   beta   logsdu   logsd0

## Profile wrt. sigma0:
prof <- tmbprofile(obj,"logsd0")
plot(prof)
confint(prof)

## Profile the difference between the beta parameters (name is optional):
prof2 <- tmbprofile(obj,name="beta1 - beta2",lincomb = c(1,-1,0,0))
plot(prof2)
confint(prof2)
} # }