A Review of Meta-Analysis Packages in R

Joshua R. Polanin & Emily A. Hennessy
July 1, 2015

Presentation Overview

  • Brief overview of meta-analysis software
  • Finding meta-analytic R packages (methods)
  • Investigating package capabilities (results)
  • Conclusions

Meta-analysis

Meta-analysis

Brief overview of meta-analysis software

Early days

Excel Meta-Analysis

Brief overview of meta-analysis software

Standard software

SPSS Meta-Analysis

SAS Meta-Analysis

Brief overview of meta-analysis software

Comprehensive meta-analysis

Comprehensive Meta-Analysis

CMA Spreadsheet

The revolution: R packages for meta-analysis

R package documentation/help

  • Traditional online sources fall short
    • Stack overflow: 85 (!) results
    • R-bloggers: 50 results (most recent: July 2014!)

Project overview

  • Question What types of meta-analytic R packages are available and what are their capabilities?

  • To answer that question:

    • Find ALL meta-analysis R packages
    • Summarize their capabilities
    • Suggest future packages and functions
    • Document help for future tutorials

Finding meta-analytic R packages: Search and screen

  • Searched multiple online repositories
    • CRAN
    • CRANtastic
    • Revolution R
  • Screened packages for inclusion
  • “Coded” (i.e., documented information) package capabilities

Finding meta-analytic R packages: Coding process

  • What we coded?
    • Name, authors, date created, date updated, pdf location
    • Package specificity (i.e., general or specific), package type (i.e., bayesian, genome, general, plotting, etc.), dependencies
    • “Traditional” meta-analytic capabilities (effect sizes, fixed or random-effects models, moderator analyses)
    • Newer meta-analytic capabilities (power analyses, missing data, handles dependencies, publication biases)
    • Main function name
    • Other capabilities
    • Package downloads

What we found: Results

Package characteristics

  • N = 56
  • All downloadable from CRAN
  • Most updated recently
  • Very wide range of capabilities and functionality
  • Yet to find feature not available in one of these packages
  • Package characteristics table

What we found: Results

When were packages created?

plot of chunk unnamed-chunk-1

What we found: Results

How often are they used?

Downloads per day (May 2015)

Source: local data frame [2 x 3]

  downloads ADD_Mean    ADD_SD
1      High 65.00806 22.984007
2       Low 10.29839  5.652116

High: n = 4 (metafor, rmeta, meta, epiR)

Low: n = 52 (all others)

What we found: Results

How often are they used?

Downloads since Oct 2012

plot of chunk unnamed-chunk-3

Top packages

  • metafor (ADD = 93.65)
  • rmeta (ADD = 50.22)
  • epiR (ADD = 49.74)
  • meta (ADD = 48.39)

What we found: Results

How often are they used? (Compared to top packages)

Downloads per day (May 2015)

Source: local data frame [2 x 3]

  meta   ADD_Mean    ADD_SD
1   No 4186.50645 795.02563
2  Yes   14.20622  16.14225
  • Top downloaded packages
 [1] "Rcpp"       "stringr"    "stringi"    "plyr"       "ggplot2"   
 [6] "magrittr"   "colorspace" "scales"     "digest"     "reshape2"  

What we found: Results

What types of packages are available?

plot of chunk unnamed-chunk-6

What we found: Results

What can the packages do?

plot of chunk unnamed-chunk-7

What we need from the R community

  • Power analyses packages or functions
  • Greater plotting capabilities (most use metafor)
  • Effect size calculations for non-traditional measures
  • Documentation!

Conclusions

  • Great capabilities
  • Straightforward to use if comfortable with R
  • Continued documentation and help will increase usage and application

Thank You!