Wrapper function to compute bootstrap analyses, leave-one-out analyses, and cumulative meta-analyses. This function helps researchers to examine the stability/fragility of their meta-analytic results with bootstrapping and leave-one-out analyses, as well as detect initial evidence of publication bias with cumulative meta-analyses.
sensitivity(
ma_obj,
leave1out = TRUE,
bootstrap = TRUE,
cumulative = TRUE,
sort_method = c("weight", "n", "inv_var"),
boot_iter = 1000,
boot_conf_level = 0.95,
boot_ci_type = c("bca", "norm", "basic", "stud", "perc"),
...
)sensitivity_bootstrap(
ma_obj,
boot_iter = 1000,
boot_conf_level = 0.95,
boot_ci_type = c("bca", "norm", "basic", "stud", "perc"),
...
)
sensitivity_cumulative(ma_obj, sort_method = c("weight", "n", "inv_var"), ...)
sensitivity_leave1out(ma_obj, ...)
An updated meta-analysis object with sensitivity analyses added.
When bootstrapping is performed, the bootstrap section of the follow_up_analyses section of the updated ma_obj returned by this function will contain both a matrix summarizing the mean, variance, and confidence intervals of the bootstrapped samples and a table of meta-analytic results from all bootstrapped samples.
When leave-one-out analyses are performed, the ma_obj will acquire a list of leave-one-out results in its follow_up_analyses section that contains a table of all leave-one-out meta-analyses along with plots of the mean and residual variance of the effect sizes in the meta-analyses.
When cumulative meta-analysis is performed, the ma_obj will acquire a list of cumulative meta-analysis results in its follow_up_analyses section that contains a table of all meta-analyses computed along with plots of the mean and residual variance of the effect sizes in the meta-analyses, sorted by the order in which studies were added to the meta-analysis.
Meta-analysis object.
Logical scalar determining whether to compute leave-one-out analyses (TRUE) or not (FALSE).
Logical scalar determining whether bootstrapping is to be performed (TRUE) or not (FALSE).
Logical scalar determining whether a cumulative meta-analysis is to be computed (TRUE) or not (FALSE).
Method to sort samples in the cumulative meta-analysis. Options are "weight" to sort by weight (default), "n" to sort by sample size, and "inv_var" to sort by inverse variance.
Number of bootstrap iterations to be computed.
Width of confidence intervals to be constructed for all bootstrapped statistics.
Type of bootstrapped confidence interval. Options are "bca", "norm", "basic", "stud", and "perc" (these are "type" options from the boot::boot.ci function). Default is "bca". Note: If you have too few iterations, the "bca" method will not work and you will need to either increase the iterations or choose a different method.
Additional arguments.
if (FALSE) {
## Run a meta-analysis using simulated correlation data:
ma_obj <- ma_r_ic(rxyi = rxyi, n = n, rxx = rxxi, ryy = ryyi, ux = ux,
correct_rr_y = FALSE, data = data_r_uvirr)
ma_obj <- ma_r_ad(ma_obj, correct_rr_y = FALSE)
## Pass the meta-analysis object to the sensitivity() function:
ma_obj <- sensitivity(ma_obj = ma_obj, boot_iter = 10,
boot_ci_type = "norm", sort_method = "inv_var")
## Examine the tables and plots produced for the IC meta-analysis:
ma_obj$bootstrap[[1]]$barebones
ma_obj$bootstrap[[1]]$individual_correction$true_score
ma_obj$leave1out[[1]]$individual_correction$true_score
ma_obj$cumulative[[1]]$individual_correction$true_score
## Examine the tables and plots produced for the AD meta-analysis:
ma_obj$bootstrap[[1]]$artifact_distribution$true_score
ma_obj$leave1out[[1]]$artifact_distribution$true_score
ma_obj$cumulative[[1]]$artifact_distribution$true_score
## Run a meta-analysis using simulated d-value data:
ma_obj <- ma_d_ic(d = d, n1 = n1, n2 = n2, ryy = ryyi,
data = filter(data_d_meas_multi, construct == "Y"))
ma_obj <- ma_d_ad(ma_obj)
## Pass the meta-analysis object to the sensitivity() function:
ma_obj <- sensitivity(ma_obj = ma_obj, boot_iter = 10,
boot_ci_type = "norm", sort_method = "inv_var")
## Examine the tables and plots produced for the IC meta-analysis:
ma_obj$bootstrap[[1]]$barebones
ma_obj$bootstrap[[1]]$individual_correction$latentGroup_latentY
ma_obj$leave1out[[1]]$individual_correction$latentGroup_latentY
ma_obj$cumulative[[1]]$individual_correction$latentGroup_latentY
## Examine the tables and plots produced for the AD meta-analysis:
ma_obj$bootstrap[[1]]$artifact_distribution$latentGroup_latentY
ma_obj$leave1out[[1]]$artifact_distribution$latentGroup_latentY
ma_obj$cumulative[[1]]$artifact_distribution$latentGroup_latentY
}
Run the code above in your browser using DataLab