logLik.pvlrt {pvLRT} | R Documentation |
Overall Log-likelihood for a pvlrt object
Description
Overall Log-likelihood for a pvlrt object
Usage
## S3 method for class 'pvlrt'
logLik(object, type = "full-zip", ...)
Arguments
object |
a |
type |
Type of model and hypothesis combination. Available choices are "full-poisson", "null-poisson", "full-zip" (default), and "null-zip". See details. |
... |
other input parameters. Currently unused. |
Details
The function extracts the overall log-likelihood and degrees of freedom
(the number of estimated parameters) from a pvlrt
object. There are
four possible choices of distribution and hypothesis combinations, supplied
through the argument type
, with the default being type = "full-zip"
.
In a "full" model the signal parameters lambdas are estimated for all cells
in the contingency table from the data (subject to the condition lambda >= 1), whereas under a "null"
model each lambda is fixed at 1 for each cell. In a "zip" model
(type = "full-zip" and type = "null-zip") the log-likelihood under a zero-inflated
Poisson model with estimated or supplied zero inflation parameters (
through zi_prob
in pvlrt) is returned. The degrees of freedom
reflects whether the zero-inflation parameters are estimated. Note that if
an ordinary Poisson LRT is run either by setting zi_prob = 0
in
pvlrt or equivalently through lrt_poisson then "full-zip" and
"null-zip" refer to zero-inflated poisson models with presepecified
zero-inflation probabilities equal to 0. Thus, in such cases the results
with type = "full-zip" and type = "null-zip" are identical to
type = "full-poisson" and type = "null-poisson"
respectively. See examples.
Value
An object of class logLik. See Details.
Note
The overall log likelihood must be computed during the original pvlrt run. This is
ensured by setting return_overall_loglik = TRUE
, and
parametrization = "lambda"
(or parametrization = "rrr"
) while running pvlrt().
See Also
Examples
# 500 bootstrap iterations (nsim) in each example below
# are for quick demonstration only --
# we recommended setting nsim to 10000 (default) or bigger
set.seed(100)
# estimates zero inflation probabilities
test1 <- pvlrt(statin46, nsim = 500)
logLik(test1)
AIC(test1)
BIC(test1)
# compare with and without zero inflation
BIC(logLik(test1, type = "full-zip"))
BIC(logLik(test1, type = "full-poisson"))
# ordinary poisson model
## equivalent to pvlrt(statin46, zi_prob = 0, nsim = 500)
test2 <- lrt_poisson(statin46, nsim = 500)
all.equal(logLik(test2, "full-zip"), logLik(test2, "full-poisson"))