- formula
an object of class formula specifying the frontier:
a typical model is y ~ x1 + ..., where y is the
log of output (or total cost), and x's are inputs (or outputs and
input prices, in logs). See Details.
- data
an optional data.frame containing the variables in formula.
If not found in data, variables are taken from environment(formula).
- subset
an optional logical or numeric vector specifying a subset of observations
for which the model is estimated and efficiencies are computed.
- distribution
character scalar specifying the distribution of the inefficiency term: default "e" (exponential).
"h" (half-normal) and "t" (truncated normal) to be implemented.
- prod
logical. If TRUE, estimates correspond to a stochastic production
frontier and technical efficiencies are returned; if FALSE, estimates
correspond to a stochastic cost frontier and cost efficiencies are returned.
Default is TRUE.
- start.val
optional numeric vector of starting values for the optimizer.
- init.sk
numeric. Initial value for the skewness parameter of the noise component;
default is 0.5.
- ln.var.u
optional one-sided formula; e.g. ln.var.u = ~ z3 + z4. Specifies exogenous
variables entering the (log) variance of the inefficiency component. If
NULL, the inefficiency variance is homoskedastic, i.e.,
\(\sigma_{u0}^2 = \exp(\gamma_{u0}[0])\).
- ln.var.v
optional one-sided formula; e.g. ln.var.v = ~ z1 + z2. Specifies exogenous
variables entering the (log) variance of the random noise component. If
NULL, the noise variance is homoskedastic, i.e.,
\(\sigma_{v0}^2 = \exp(\gamma_{v0}[0])\).
- skew.v
optional one-sided formula; e.g. skew.v = ~ z5 + z6. Allows the skewness
of the noise to depend linearly on exogenous variables. If NULL, the
skewness is constant across units.
- mean.u
optional one-sided formula; e.g. mean.u = ~ z7 + z8. Specifies whether the
mean of the pre-truncated normal distribution of the inefficiency term is a
linear function of exogenous variables. In cross-sectional models, used only
when distribution = "t". If NULL, the mean is constant across units. To be implemented.
- technique
Optimization technique to use.
- vcetype
Type of variance-covariance matrix estimation.
- optim.method
character. Method passed to stats::optim when optim = TRUE.
Default is "bfgs".
- optim.report
integer. Verbosity level for reporting during optimization (if implemented).
Default is 1.
- optim.trace
integer. Trace level for optimization (if implemented). Default is 1.
- reltol
numeric. Relative convergence tolerance used when maximizing the log-likelihood.
- optim.reltol
numeric. Relative tolerance specifically for optim; default 1e-8.
- lmtol
numeric. Convergence tolerance based on the scaled gradient (when applicable).
Default is 1e-5.
- maxit
numeric. Maximum number of iterations for the optimizer. Default is 199.
- print.level
integer. Printing level: 1—estimation results;
2—optimization details; 3—summary of (cost/technical)
efficiencies; 4—unit-specific point and interval estimates of
efficiencies. Default is 0.
- threads
Number of threads for parallel computation.
- only.data
Logical; if TRUE, return only processed data.
- digits
integer. Number of digits for displaying estimates and efficiencies. Default is 4.
- ...
Additional arguments (currently unused).
- optim
logical. If TRUE, estimation proceeds via stats::optim; if
FALSE, an internal routine (if provided) would be used. Default is FALSE.
- report
Reporting level for optimization progress.
- trace
Logical; if TRUE, trace optimization progress.