ribbonROE {bayesROE}

Bayesian Regions of Evidence Ribbon Plot

Description

Compute and visualize the Bayesian Regions of Evidence (Ribbon), that is, the set of normal priors for an effect size which - when combined with the observed data - lead to a specified posterior probability for the effect size being more extreme than a specified minimally relevant effect size.

Usage

ribbonROE(
  ee,
  se,
  delta = 0,
  alpha = 0.025,
  type = "threshold",
  larger = TRUE,
  meanLim = c(pmin(2 * ee, 0), pmax(0, 2 * ee)),
  sdLim = c(0, 3 * se),
  nGrid = 500,
  relative = TRUE,
  cols = NULL,
  cols_alpha = 1,
  addRef = TRUE,
  sceptPrior = 0,
  addEst = FALSE
)

Arguments

ee

Effect estimate.

se

Standard error of effect estimate.

delta

Minimally relevant effect size. Defaults to zero. Can also be a vector of numerical values to representing different regions.

alpha

Posterior probability that the effect size is less extreme than delta. Defaults to 0.025. Can also be a vector of numerical values representing different regions.

type

Character indicating if regions of evidence should be constructed for a non-inferiority claim using the first element of delta and all elements of alpha ("threshold") or for a non-inferiority claim using the all elements of delta and the first element of alpha ("probability"). Defaults to "threshold".

larger

Logical indicating if effect size should be larger (TRUE) or smaller (FALSE) than delta. Defaults to TRUE.

meanLim

Limits of prior mean axis. Defaults to interval between zero and two times the effect estimate.

sdLim

Limits of prior standard deviation axis. Defaults to interval between zero and three times the standard error.

nGrid

Number of grid points (on the standard error axis). Defaults to 500.

relative

Logical indicating whether a second x-axis and y-axis with relative prior mean and relative prior variance should be displayed. Defaults to TRUE.

cols

Character containing the HEX color code of the upper and lower region of evidence, respectively. Defaults to NULL, which triggers automated color picking by calling ggplot2:scale_fill_viridis_d()

cols_alpha

Numeric value indicating the relative opacity of any region of evidence (alpha channel). Defaults to 1 (no transparency).

addRef

Logical indicating if a reference cross representing the minimum sceptical prior is added to the plot. Defaults to TRUE.

sceptPrior

Numeric value indicating the effect location of the minimum sceptical prior. Defaults to 0.

addEst

Logical indicating if a point symbol representing the mean and standard error of the effect estimate (ee, se) is added to the plot. Defaults to FALSE.

Value

A bayesROE object (a list containing the ggplot object, the data for the plot, and the tipping point function)

References

Pawel, S., Matthews, R. and Held, L. (2021). Comment on "Bayesian additional evidence for decision making under small sample uncertainty". Manuscript submitted for publication. Code available at https://osf.io/ymx92/

Examples

## data with p < 0.025 for H0: delta < 0, but p > 0.025 for H0: delta < 0.3
d <- 0.4
d_se <- 0.1
delta <- c(0, 0.3)
ribbonROE(ee = d, se = d_se, delta = delta, meanLim = c(-1, 1))

## reproducing Figure 1 from Hoefler & Miller (2023)
ee <- 3.07
se <- 1.19
ribbonROE(ee = ee, se = se, delta = c(0,3), alpha = 0.025, 
  cols = c("#F5FF82", "#27CC1E"))$plot + 
  ggplot2::annotate(geom = "point", y = ee, x = se, shape = 4) +
  ggplot2::coord_flip(ylim = c(-5, 15))