In dsquintana/metameta: A Suite of Tools for calculating and visualising study-level statistical power for meta-analyses

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.width = 7,
  fig.height = 3
)

The metameta package includes a set of functions to re-analyse published meta-analyses. Let's begin by calculating the statistical power of each study included in a meta-analysis. This function will calulate power assuming that the reported summary effect size estimate is the true effect sizes (the "observed_es" argument) as well as assuming a range of possible true effect sizes, ranging from 0.1 to 1 (in increments of 0.1).

This particular meta-analysis includes 9 studies, and reported both the effect size and standard error in a forest plot (for more information on this study, see the documentation page: ?dat_ooi). Here, we're only showing the first 6 columns for the purposes of illustration, although the function calculates power for 11 true effect sizes in total.

library(metameta)
library(dplyr)

power_ooi <- mapower_se(dat = dat_ooi, observed_es = 0.178, name = "ooi et al 2017")
power_ooi_dat <- power_ooi$dat
power_ooi_dat <- power_ooi_dat %>% select(1:6) # Select first 6 columns
power_ooi_dat

There's also a similar function for calculating the statisical power of a meta-analysis that reports effect size and confidence interval data, which is often reported if standard error data isn't reported. As before, we're only printing a selection of columns.

library(metameta)
library(dplyr)

power_keech <- mapower_ul(dat = dat_keech, observed_es = 0.08, name = "Keech et al 2017")
power_keech_dat <- power_keech$dat
power_keech_dat <- power_keech_dat %>% select(1:6) # Select first 6 columns
power_keech_dat

Sometimes it's useful to calculate the statistical power for a body of meta-analyses, which might be reported in the same article or accross articles. Illustrating the power of individual studies from multiple meta-analyses can be difficult to interpret if there are many studies. An alternative is to illustrate the power per meta-analysis by calculating the mean power accross studies. We can illustrate this with a "Firepower" plot.

Before we create our Firepower plot, we need to prepare the datafile

### Calcuate median power for three meta-analyses
power_ooi <- mapower_se(dat = dat_ooi, observed_es = 0.178, name = "ooi et al 2017")
power_med_ooi <- power_ooi$power_median_dat

keech_power <- mapower_ul(dat = dat_keech, observed_es = 0.08, name = "Keech et al 2017")
power_med_keech <- keech_power$power_median_dat

power_bakermans_kranenburg <- 
  mapower_se(dat = dat_bakermans_kranenburg, observed_es = 0.32, 
             name = "Bakermans-Kranenburg et al 2013")
power_med_bakermans_kranenburg <- power_bakermans_kranenburg$power_median_dat

### Create a list
list_power <- list(power_med_ooi, power_med_keech, power_med_bakermans_kranenburg)

Now, we can create our plot using the list we created.

### Run firepower function
fp <- firepower(list_power)

### Create firepower plot
fp_plot <- fp$fp_plot
fp_plot

You can also print and store the data underlying this figure.

fp_data <- fp$dat
fp_data

dsquintana/metameta documentation built on Oct. 14, 2022, 2:16 a.m.