compare_posterior

Source: R/compare_posterior.R

Extracts posterior predicted values from a list of class bayesnecfit or bayesmanecfit model fits and compares these via bootstrap re sampling.

compare_posterior(
  x,
  comparison = "n(s)ec",
  ecx_val = 10,
  type = "absolute",
  hormesis_def = "control",
  sig_val = 0.01,
  resolution,
  x_range = NA,
  make_newdata = TRUE,
  ...
)

Arguments

x

A named list of objects of class bayesnecfit or bayesmanecfit returned by bnec.

comparison

The posterior predictions to compare, takes values of "nec", "n(s)ec", "nsec", "ecx" or "fitted".

ecx_val

The desired percentage effect value. This must be a value between 1 and 99 (for type = "relative" and "absolute"), defaults to 10.

type

A character vector, taking values of "relative", "absolute" (the default) or "direct". See Details.

hormesis_def

A character vector, taking values of "max" or "control". See Details.

sig_val

Probability value to use as the lower quantile to test significance of the predicted posterior values.

resolution

The number of unique x values over which to find ECx – large values will make the ECx estimate more precise.

x_range

A range of x values over which to consider extracting ECx.

make_newdata

Only used if comparison = "fitted". Should the user allow the package to create newdata for predictions? If so, arguments resolution and x_range will be used. Defaults to TRUE. See details.

...

Further arguments that control posterior predictions via posterior_epred.

Value

A named list containing bootstrapped differences in posterior predictions of the bayesnecfit or bayesnecfit model fits contained in x. See Details.

Details

type "relative" is calculated as the percentage decrease from the maximum predicted value of the response (top) to the minimum predicted value of the response. Type "absolute" (the default) is calculated as the percentage decrease from the maximum value of the response (top) to 0 (or bot for a 4 parameter model fit). Type "direct" provides a direct estimate of the x value for a given y. Note that for the current version, ECx for an "nechorme" (NEC Hormesis) model is estimated at a percent decline from the control.

For hormesis_def, if "max", then ECx or NSEC values – i.e., depending on argument comparison – are calculated as a decline from the maximum estimates (i.e. the peak at NEC); if "control", then ECx or NSEC values are calculated relative to the control, which is assumed to be the lowest observed concentration.

The argument make_newdata is only used if comparison = "fitted". It is relevant to those who want the package to create a data.frame from which to make predictions. This is done via bnec_newdata and uses arguments resolution and x_range. If make_newdata = FALSE and no additional newdata argument is provided (via ...), then the predictions are made for the raw data. Else, to generate predictions for a specific user-specific data.frame, set make_newdata = FALSE and provide an additional data.frame via the newdata argument. For guidance on how to structure newdata, see for example posterior_epred.

Examples

if (FALSE) { # \dontrun{
library(bayesnec)
data(manec_example)
nec4param <- pull_out(manec_example, model = "nec4param")
ecx4param <- pull_out(manec_example, model = "ecx4param")
compare_posterior(list("n(s)ec" = ecx4param, "ecx" = nec4param), ecx_val = 50)
} # }