ll.call: Call a likelihood/link-specific function

Description

GeMTC implements various likelihood/link combinations. Functionality specific to the likelihood/link is handled by methods with names ending in .<likelihood>.<link>. This convenience function calls such methods.

Usage

ll.call(fnName, model, ...)

Arguments

fnName

The name of the function to call. See details for available functions.

model

An object of S3 class mtc.model describing a network meta-analysis model, or a list containing elements named 'likelihood' and 'link'.

...

Additional arguments to be passed to the function.

Value

The return value of the called function.

Details

The following methods currently need to be implemented to implement a likelihood/link:

mtc.arm.mle: calculates a (corrected) maximum likelihood estimate for an arm-level effect. Used to generate starting values.
mtc.rel.mle: calculates a (corrected) maximum likelihood estimate for a relative effect. Used to generate starting values.
mtc.code.likelihood: generates JAGS code implementing the likelihood.
scale.log: returns TRUE if plots should use the log scale.
scale.name: returns the user-facing name of the outcome metric.
scale.limit.inits: returns an upper and lower bound for the initial values, because some initial values might trigger boundary conditions such as probability 0 or 1 for the binomial.
required.columns.ab: returns the required columns for arm-based data.

The first two methods can now also be used to selectively apply continuity corrections in case the maximum likelihood estimates are used for other purposes. mtc.arm.mle has an additional k=0.5 argument to specify the correction factor. mtc.rel.mle has arguments correction.force=TRUE to force application of the continuity correction even if unnecessary, correction.type="constant" to specify the type of correction (specify "reciprocal") for a correction proportional to the reciprocal of the size of the other arm, and correction.magnitude=1 to specify the (total) magnitude of the correction. These corrections apply only for count data, and will be ignored for continuous likelihood/links.

Examples

Run this code

# NOT RUN {
# The "model" may be a stub.
model <- list(likelihood="poisson", link="log")

ll.call("scale.name", model)
# "Hazard Ratio"

ll.call("mtc.arm.mle", model, c('responders'=12, 'exposure'=80))
#      mean         sd
#-1.8562980  0.1118034
# }