Skip to contents

ECDF Plot Comparing Model Strategies

Source: R/ecdf_model_plot.R
ecdf_model_plot.Rd

Creates an empirical cumulative distribution function (ECDF) plot comparing posterior distributions across different model strategies (e.g., random effects, selection models, PET-PEESE) under a single prior specification.

Use this plot to assess how robust conclusions are across different modelling assumptions.

Usage

ecdf_model_plot(
  model,
  data,
  prior,
  estimand,
  rob_var = NULL,
  exclude_high_rob = FALSE,
  incl_common_effect = FALSE,
  incl_random_effect = TRUE,
  incl_bias_corrected = FALSE,
  incl_selection_copas = FALSE,
  incl_selection_weight = FALSE,
  incl_pet_peese = FALSE,
  incl_robust = FALSE,
  incl_robma = FALSE,
  prob_reference = "null",
  null_value = NULL,
  null_range = NULL,
  add_null_range = FALSE,
  color_null_range = "#77bb41",
  label_control = "Control",
  label_intervention = "Intervention",
  title = NULL,
  subtitle = NULL,
  xlim = NULL,
  x_breaks = NULL,
  color_palette = NULL,
  show_density = TRUE,
  font = NULL
)

Arguments

model

A fitted bayesma object.

data

A data frame containing the study data used to fit the model.

prior

A list containing the prior specification with elements:

mu_prior

(Required) Prior specification for mu.

tau_prior

(Optional) Prior specification for tau.

name

Display label for the prior. Required when incl_robma = TRUE as it is used to match against priors in model$robma_sensitivity. If not provided and incl_robma = FALSE, defaults to "User-specified prior".

estimand

Effect estimand string (e.g., "OR", "RR", "HR", "IRR", "MD", "SMD").

rob_var

Optional. Name of the risk-of-bias variable (unquoted).

exclude_high_rob

Logical. If TRUE, includes an "Excluding High RoB" section.

incl_common_effect

Logical. Include common effect model. Default FALSE.

incl_random_effect

Logical. Include random effects model. Default TRUE.

incl_bias_corrected

Logical. Include bias-corrected model. Default FALSE.

incl_selection_copas

Logical. Include Copas selection model. Default FALSE.

incl_selection_weight

Logical. Include weight-function selection model. Default FALSE.

incl_pet_peese

Logical. Include PET-PEESE model. Default FALSE.

incl_robust

Logical. Include robust mixture model. Default FALSE.

incl_robma

Logical. Include RoBMA model. Requires model$robma_sensitivity to be present and prior$name to match one of the priors used when fitting. Default FALSE.

prob_reference

Character string. What reference to use for probability axis labels. Either "null" (compare to null_value) or "null_range" (compare to null_range boundaries). Default is "null".

null_value

Null hypothesis value. If NULL, uses estimand default.

null_range

Numeric vector of length 2 giving null range bounds.

add_null_range

Logical. If TRUE and null_range is NULL, uses estimand-appropriate defaults.

color_null_range

Fill colour for the null range band. Default "#77bb41".

label_control

Label for control group. Default "Control".

label_intervention

Label for intervention group. Default "Intervention".

title

Optional plot title.

subtitle

Optional plot subtitle. If NULL and a prior name is available, displays the prior name.

xlim

Optional x-axis limits.

x_breaks

Optional x-axis break points.

color_palette

Optional named vector of colours for model sections.

show_density

Logical. If TRUE, includes a density plot below the ECDF. Default is TRUE.

font

Optional font family.

Value

A ggplot object (or patchwork object if show_density = TRUE).

Details

The plot shows one ECDF line per model strategy, all using the same prior. This allows direct comparison of how different modelling assumptions (random effects, selection models, bias correction, etc.) affect the posterior distribution.

The left y-axis shows P(effect < x) and the right y-axis shows P(effect > x).

See also

ecdf_prior_plot for comparing priors within a model type.

Examples

if (FALSE) { # \dontrun{
priors <- list(
  vague = list(
    name = "Vague",
    mu_prior = normal(0, 10),
    tau_prior = half_cauchy(0, 1)
  ),
  weak_reg = list(
    name = "Weakly Regularising",
    mu_prior = normal(0, 1),
    tau_prior = half_cauchy(0, 0.5)
  )
)

ecdf_model_plot(
  model = model,
  data = dat,
  estimand = "OR",
  prior = priors$weak_reg,
  incl_random_effect = TRUE,
  incl_pet_peese = TRUE
)

ecdf_model_plot(
  model = model,
  data = dat,
  estimand = "OR",
  prior = priors$weak_reg,
  incl_robma = TRUE
)
} # }