Tools for computing bare-bones and psychometric meta-analyses and for generating psychometric data for use in meta-analysis simulations. Supports bare-bones, individual-correction, and artifact-distribution methods for meta-analyzing correlations and d values. Includes tools for converting effect sizes, computing sporadic artifact corrections, reshaping meta-analytic databases, computing multivariate corrections for range variation, and more. Bugs can be reported to < https://github.com/psychmeta/psychmeta/issues> or
The psychmeta
package provides tools for computing bare-bones and psychometric meta-analyses and for generating psychometric data for use in meta-analysis simulations. Currently, the package supports bare-bones, individual-correction, and artifact-distribution methods for meta-analyzing correlations and d values. Please refer to the overview tutorial vignette for an introduction to psychmeta
's functions and workflows.
psychmeta
was written by Jeffrey A. Dahlke and Brenton M. Wiernik.
The official CRAN release of psychmeta
can be installed with the following code:
install.packages("psychmeta")
Development versions of psychmeta
from GitHub reflect updates made to the package between official CRAN releases. Using the devtools package, the GitHub release can be installed with the following code:
install.packages("devtools")devtools::install_github("psychmeta/psychmeta")
psychmeta
To cite psychmeta
in your research, please refer to the package's citation information using the citation()
function.
citation("psychmeta")
To report a bug or other issue, tell us about it on GitHub or email [email protected]. For more general questions and inquiries about the package, reach out to us via Twitter or email [email protected].
Bug fix for using categorical moderators in meta-regressions.
Bug fix for using moderators in ma_generic() when there are no construct or group-identification variables in the analysis.
Added an optional "weights" argument to ma_generic(), which allows users to supply their own custom weights for effect sizes.
Added alternative methods for calculating confidence limits for tau-squared.
Increased support for heterogeneity analyses conducted using weights imported from the metafor package.
Updated estimates for small-sample-bias corrected error variances to be computed from the bias-corrected effect size rather than applying a post-hoc linear variance adjustment.
Changed default weighting method for d values to "n_effective." This method is superior to sample-size weights because it accounts for both total sample size and deviations from equal subgroup representation.
Updated compatibility with dplyr version 0.8.0.
Updated the labeling of output from the mix_dist function.
Assorted bug fixes affecting mean d value estimates (from the ma_d and ma_d_ic functions), artifact distributions, and weighted correlations and covariances.
Individual-correction validity generalization meta-analyses have been corrected to remove variation due to between-sample reliability differences prior to adding back measurement error.
Improved compatibility with dplyr and tibble packages.
Bug fixes for var_error_d
and var_error_g
to accommodate inputs of differing lengths.
Bug fix for cumulative meta-analyses.
Bug fix for handling null interactive ADs
Bug fix for meta-analyzing correlations between global constructs and facets.
Bug fix for estimating correction-implied subgroup proportions in correct_d.
Bug fix for using correct_d when n2 is left unspecified.
Added option to suppress progress indicators in the console. Use options(psychmeta.show_progress = FALSE) to suppress progress indicators.
Increased support for common wide database formats in the reshape_wide2long() function.
Bug fixes for managing dependent observations in artifact distributions created within ma_r() and ma_d().
Bug fixes for resolving dependency in effect sizes in meta-analyses with moderators.
Added error-variance estimates to the output of convert_es() for increased compatibility with the ma_generic() function.
Added support for separately compositing measures of constructs and the constructs' facets. Arguments for 'facet_x' and 'facet_y' have been added to ma_r() and ma_d() so that more comprehensive output tables can be generated.
Stability fixes for data management within the ma_r() function.
Added checks for missing and infinite values to the ma_generic() function.
Bug fix for create_ad() with interactive method when all sample sizes are NA.
Bug fix for ma_r_ad() with individual artifact distributions supplied.
Bug fix for using output of metabulate() in Rmarkdown documents.
Bug fix for meta-analyses of effect sizes that have been averaged or composited.
Increased the stability of artifact-distribution meta-analyses with moderators computed using the ma_r() function.
Bug fix for the get_ad() function to appropriately handle artifact-distribution objects that contain no artifact information.
Updated and corrected the text displayed in table notes produced by the metabulate() function.
Bug fix for handling meta-analyses with only one effect size.
Removed package 'fungible' as a dependency, as 'fungible' was archived and removed from CRAN on 2018-07-15.
Bug fixes and improvements to format_num() and metabulate() for handling leading zeros.
Changed unicode entities to HTML entities for increased support across locales.
Bug fix for artifact imputation without a moderator.
Added grouping variables and construct variables to ma_generic(). This helps to streamline meta-analyses of generic effect sizes that involve multiple group-wise contrasts and/or multiple constructs.
Bug fixes for format_num() and metabulate() to allow missing values in formatted numeric strings.
Updated lm_mat()to use fungible::seBetaCor to estimate standard errors when se_beta_method is "normal" and cov_mat is a correlation matrix.
Updates and bug fixes for the metabulate() function. Fixed issues with file-writing protocols. Increased support for unicode characters on Windows systems. Added arguments to control whether table headers are rendered in bolded font (bold_headers) and whether redundant construct-pair labels are collapsed in meta-analysis tables (collapse_construct_labels).
Updated the “summary” attributes of artifact-distribution objects for greater consistency between interactive and Taylor-series objects.
Added third-order sample sizes (i.e., the number of meta-analyses meta-analyzed; denoted as “L”) to the output from second-order meta-analyses.
Added get_stuff() function that serves as a wrapper for other functions in the “get_stuff” family of functions.
Updated "metabulate" function based on RMarkdown. The metabulate function can now output to Word, PDF, HTML, Markdown, ODT, and plain text and accepts arguments to include a bibliography in the same output as results tables. Output can be rendered directly into the selected format or placed into a larger RMarkdown document. The function also includes improvements to the typesetting of numeric results and column headings (e.g., ρ̅ instead of Mean_{ρ}).
New "num_format" function to format numbers to a specified number of digits and allowing control of the characters used for decimal points, thousands and decimal digit group separators, leading zeros, and positive and negative signs. It's useful to avoid unpleasant surprises during publisher typesetting of tables.
Hotfix for handling NULL moderator objects in dependency resolution procedure.
Hotfix for missing object in the dependency resolution procedure when no moderators are present.
Major overhauls to the format of meta-analysis objects to improve the user experience. The outputs of meta-analysis functions are now nested tibbles rather than nested lists. This provides users with a clearer overview of the structure of the object, streamlines the manipulation of results objects (e.g., selecting/subsetting columns, filtering, and arranging), and makes the objects more flexible in terms of performing follow up analyses and performing artifact-distribution corrections (because AD corrections can now by done separately by moderator level). As the print methods for meta-analyses no longer report grand tables of results, we recommend using summary() or get_metatab() to obtain this information.
Arguments for specifying methodological parameters (e.g., conf_method, conf_level, var_unbiased) in meta-analyses have been migrated to a new control_psychmeta() function - the output of this function can be passed to the “control” argument of meta-analysis functions to select the methodological parameters. This is meant to reduce the number of arguments native to meta-analysis functions and simplify the documentation of psychmeta’s core functions. All of the arguments to control_psychmeta can still be passed directly to meta-analysis functions and all function calls accepted by previous versions of psychmeta should continue to work in this version of the package.
The dependency resolution process in ma_r() and ma_d() in which dependent effect sizes are automatically combined into composites has been enhanced. The “intercor” argument now accepts a variety of types of information. The user can now set a default scalar value to use as well as supply a named vector of values (the names can refer to construct names and/or sample_id and construct pairings) and/or a database from which to harvest intercorrelation information. This allows the compositing procedure to use sample-specific information to compute composites when that information is available.
“summary” methods have been added for classes exported from psychmeta.
The overview vignette has been updated to provide instructions on how to navigate the output objects from meta-analyses.
The class structure of meta-analysis objects has been simplified. There is now only one class for meta-analyses ("ma_psychmeta").
Data management has been streamlined within the ma_r() and ma_d() functions to permit more efficient cleaning and imputation of artifact information.
To complement the new structure of meta-analysis objects, the create_ad_list function now produces tibbles of artifact distributions and the function is now an alias of create_ad_tibbles. This function can now create artifact distributions for construct pairs and/or specific moderator combinations.
The behavior of the supplemental_ads argument to meta-analysis functions has been expanded to allow artifact information to be supplied as individual artifact distributions, lists of artifact distributions, or tibbles of artifact distributions.
"print" methods have been updated for a variety of object classes.
Bug fixes and stability improvements have been made to follow-up analyses, plotting functions, dependency resolution, convert_ma(), and dMod().
The performance of the "hs_override" argument has been improved in ma_r and ma_d by correcting a bug that caused the argument to not be evaluated when running an artifact-distribution meta-analysis or only a bare-bones meta-analysis.
The default value for the "use_all_arts" argument in ma_r and ma_d has been changed from FALSE to TRUE. This is meant to improve the ease with which the most robust artifact-distribution corrections can be applied.
Bug fixes have been made for defining confidence-interval method in bare-bones and artifact-distribution meta-analyses computed with the ma_r function.
A new “seed” argument has been added to ma_r and ma_d to set the seed value when imputing artifacts for greater reproducibility.
New console messages have been introduced for when follow-up results are added to a meta-analysis object.
Bug fixes for using true-score u ratios in individual-correction meta-analyses.
Bug fix for appropriately returning results from the compute_dmod function.
New “citekey” arguments have been added to meta-analysis function that allow users to supply citation keys for use in reference-management programs. The new generate_bib() function can harvest the citation keys from studies that are included in meta-analysis objects and create formatted reference lists for use in publications.
Plotting functions have been added. The plot_forest() and plot_funnel() functions use ggplot2 to create forest and funnel plots, respectively. These functions take a meta-analysis object as an input and add a list of plots to the object.
New functions to retrieve results from within meta-analysis objects have been added. The “get_stuff” family of functions provide easy ways to extract meta-analysis tables, follow-up analyses (e.g., cumulative meta-analyses, leave-one-out meta-analyses, bootstrap results), plots, and more.
Assorted bug fixes have been implemented to improve the stability of the dependency resolution process, the handling of artifact distributions, and the accommodation of continuous moderators.
A new method to account for sampling error in interactive artifact distributions has been added (NOTE: this is now the default methods used in interactive AD meta-analyses).
Stability improvements for dMod functions.
Artifacts harvested from rows of a database that do not contain effects sizes in the ma_r and ma_d functions are now organized to avoid redundancy with observations from the same sample that are included in the meta-analytic computations.
Bug fixes to metabulate().
Bug fix to the interactive artifact-distribution method for bivariate indirect range restriction (bvirr) to properly compute SD rho.
Bug fix to allow interactive artifact distributions to be rounded to the desired number of decimal places.
Bug fix for computing the mean reliability indices in the Taylor-series estimate of the corrected error variance for correlations corrected with the bivariate indirect range restriction (bvirr) correction.
A new function (called matreg, with aliases of matrixreg, lm_mat, and lm_matrix) has been added that computes regression models from covariance matrices and vectors of means and generates output that resembles that of the "lm" function. Output from matreg can be used with the summary(), predict(), and confint() functions.
The ma_r and mad_d functions have been modified to allow for more flexibility in computing multiple meta-analyses in a single function call, especially when those meta-analyses use artifact-distribution corrections. Several new arguments have been added to allow users to name the types artifact corrections they would like to apply to each construct - see function documentation for argument details.
The correct_rxx and correct_ryy arguments across meta-analysis routines (excluding ma_r_ad and ma_d_ad) can now support both scalar and vector arguments.
Efficiency improvements and improvements to the accuracy of progress-bar feedback have been made to the simulate_r_database and simulate_d_database functions.
Bug fixes and stability improvements for moderator analyses, the creation of artifact distributions, and computation of follow-up analyses.
Added reliability types arguments (e.g., “rxx_type”, “ryy_type”) to functions requiring reliability estimates to be corrected for range restriction. String vectors of reliability labels (e.g., “alpha”, “retest”, “parallel”; see documentation for the ma_r function for a complete list of supported labels) can now be supplied to many functions. When direct range restriction occurs, different corrections for range restriction are applied to internal-consistency reliability estimates and reliability estimates computed by correlating data from different testing occasions.
The ma_r and ma_d functions now allow users to supply lists of external artifact information using the “supplemental_ads” argument (the ma_r_ic and ma_d_ic functions also allow this with the “supplemental_ads_x” and “supplemental_ads_y” arguments). These functions can also now harvest artifact information from studies in a database with invalid or missing effect sizes by setting the “use_all_arts” argument to TRUE.
“rho_params” arguments to the simulate_r_database function can now be supplied as a correlation matrix rather than a list (for situations in which samples are drawn from a single population). Similarly, elements in the “rho_params” list argument for the simulate_d_database can include matrices.
Progress bars have been added to potentially time-consuming functions to provide feedback about the percentage of progress and the estimated time remaining.
Bug fixes for moderator analyses.
Bug fixes for specifying the order of constructs and selecting a subset of constructs in a database.
New function to correct for scale coarseness.
Improved vectorization support.
We have added a suite of simulation function for d values (see the "simulate_d_sample" and "simulate_d_database" functions).
Methods for meta-analyzing d values have been updated to better account for subgroup proportions when converting between the r and d metrics. The escalc objects that accompany a meta-analysis of d values now include more detailed information about incumbent and applicant subgroup proportions and the default for the "pa" argument (i.e., applicant proportions) is now NULL rather than .5 to prevent unintended corrections for range restriction.
A new "ma_generic" function has been added that allow users to do a psychmeta-style meta-analysis for any effect size for which the user has error-variance estimates. This function is supported by psychmeta's sensitivity() and heterogeneity() functions.
Taylor series methods for estimating the variance of converted artifact values (e.g., a rxxi distribution converted to a rxxa distribution) have been expanded to account for the correlation between artifact distributions when such information is available. Correlations among artifacts' sampling distributions can also be passed to the var_error_r_bvirr() function when it is called outside of a meta-analysis routine.
The implementation of corrections of bivariate range restriction in individual-correction meta-analyses have been updated. Corrections for bivariate direct range restriction now use conventional compound attenuation factors, while the attenuation factors for bivariate indirect range restriction are now computed using mean effect sizes and artifacts to estimate sampling variances.
Assorted bugs have been fixed (e.g., an error that occurred when running multi-construct meta-analyses without specifying moderators; filters to retain valid correlations did not screen construct names, sample ids, or measure names; subgroup proportions were not being retained when d values were meta-analyzed as correlations).
psychmeta has officially launched for public beta!
Please see the "psychmeta-package" entry in the psychmeta manual for an overview of the package and its applications.