Skip to contents

metaprep: Tidy Preparation of Dependent Effect Estimates for Meta-Analysis

Source: R/metaprep-package.R
metaprep-package.Rd

Meta-analyzing effect estimates from multi-arm trials means accounting for the dependence that arises when several treatment arms are compared against a shared control group. metafor::rma.mv() implements the methods: you hand it a block-diagonal variance-covariance matrix that encodes the dependence. The awkward part is building that matrix and keeping it aligned with the estimates as you filter, group, and subset them. metaprep does that bookkeeping.

Where to start

The workflow has four steps, and the whole package is organized around them.

  1. Prepare each fit. prep_fit() runs tidy(), glance(), and vcov() on a fitted model and keeps the terms you name, returning them as list-columns of a one-row tibble. Bind one per study with dplyr::bind_rows(.id = "study").

  2. Build the object. as_estimates_vcov() turns those bound fits into an estimates_vcov object: the estimates and their block-diagonal vcov, held together. When the covariances do not come from a single regression, for example bootstrapped across experiments that share subjects, build it with make_estimates_vcov() instead.

  3. Reshape it. dplyr verbs work on the object and keep the vcov aligned (see dplyr-methods). To change estimate values, use rescale_estimates_vcov(), which transforms the vcov to match. To stack objects prepared separately, use bind_estimates_vcov().

  4. Pool it. rma_mv_helper() reads the estimates and vcov straight off the object and passes them to metafor::rma.mv(), optionally with cluster-robust standard errors.

get_estimates_df(), get_glance_df(), and get_vcov() pull the components back out at any point.

What the package guards against

Two failures in this workflow are silent, and both would make a meta-analysis wrong with no visible symptom, so metaprep errors rather than guessing:

  • A vcov that cannot be a covariance matrix. Asymmetry beyond floating-point noise means rows and columns are misaligned; non-finite entries mean a rank-deficient fit returned coefficients without usable standard errors. Both are rejected when the object is built.

  • An estimate that cannot enter a pool. A non-finite estimate is rejected by rma_mv_helper() and rma_uni_helper(), because metafor would drop it silently and return a fit holding fewer estimates than the object.

In each case which estimates to drop is an analyst's decision, so the package stops and leaves it to you rather than choosing quietly.

See also

Useful links:

Author

Maintainer: Alex Coppock acoppock@gmail.com

Authors:

Examples

library(dplyr)

set.seed(1)
dat_1 <- data.frame(Y = rnorm(60), Z = factor(rep(c("T0", "T1"), each = 30)))
dat_2 <- data.frame(Y = rnorm(90), Z = factor(rep(c("T0", "T1", "T2"), each = 30)))

ev <- as_estimates_vcov(bind_rows(
  study_1 = prep_fit(lm(Y ~ Z, dat_1), term = "ZT1"),
  study_2 = prep_fit(lm(Y ~ Z, dat_2), term = c("ZT1", "ZT2")),
  .id = "study"
))
ev
#> <estimates_vcov>
#> # 3 estimates with 3x3 vcov matrix
#> 
#> # A tibble: 3 × 7
#>   id    study   term  estimate std.error statistic p.value
#>   <chr> <chr>   <chr>    <dbl>     <dbl>     <dbl>   <dbl>
#> 1 1     study_1 ZT1    0.0503      0.223    0.226   0.822 
#> 2 2     study_2 ZT1    0.00305     0.238    0.0128  0.990 
#> 3 3     study_2 ZT2   -0.440       0.238   -1.85    0.0679

# Subsetting the estimates subsets the vcov with them
dim(get_vcov(ev))
#> [1] 3 3
dim(get_vcov(filter(ev, study == "study_2")))
#> [1] 2 2