Recommends a sample size n* for a prospective study, given
an elicited prior described by a bgm_elicited object. The planning
criterion depends on method, and the methods available depend on the
class of params. See the method pages listed under Details.
Usage
design(params, method = c("DPIR", "BFDA", "BSDA"), ...)Arguments
- params
A
bgm_elicitedobject, as returned byelicit_prior.- method
Which planning method to use,
"DPIR"(default),"BFDA", or"BSDA". One method runs per call.- ...
Method-specific arguments passed to the method (e.g.,
max_n, the largest sample size considered in the planning).
Value
A bgm_design object. It records the recommended sample sizes (
two per call, depending on method) together with the criterion
values and the elicited prior the planning was built from. print and
summary methods are available.
Details
Both methods look for the smallest sample size at which a criterion reaches a
target threshold with a specified probability. They differ in the criterion:
"DPIR" uses a data-to-prior information ratio over the model
parameters, "BFDA" the Bayes factor at a representative edge.
Methods are available for the following classes:
ggm_elicitedGaussian graphical models
See also
Other sample size planning:
bsda_control(),
design.ggm_elicited(),
validate(),
validate.ggm_design()