Exploratory meta-analysis notes - IMPORTANT - READ ME!

DESCRIPTION: The exploratory meta-analysis function takes as input a single search term (5-digit code from the taxonomy) and then runs all possible meta-analyses with all other taxonomic nodes (limited to cases where kEffects >= 5). It then displays each meta-analytic summary, scaling the node size by k (either kEffects, kSamples, or kArticles) and scales the node color by effect size (either r [raw correlation] or |r| [absolutized correlation]. Finally, the nodes are plotted according to the metaBUS taxonomy's structure. This advanced visualization allows the user to visualize many meta-analytic results, often based on thousands of existing findings, in mere seconds. An important limitation: We run all possible meta-analyses in advance and store their results (around 500,000 unique taxonomic pairs). In order to save computation time, we run a bare-bones Hunter-Schmidt meta-analysis. So, exploratory analyses WILL NOT necessarily match the targeted meta-analysis results in this platform (which are based on multilevel meta-analysis). In future embodiments of this function, we will provide more stable multilevel meta-analytic estimates.
USAGE: First, enter a 5-digit taxonomic code from the taxonomy (e.g., 20072 = job satisfaction [other taxonomy codes may be found by clicking EXPLORE TAXONOMY in the left panel]). Next, select a minimum k to be plotted. The minimum k feature prevents the server from running too many nuanced analyses (PLEASE start with a large number and progress to smaller k values as needed, or the server might crash... seriously). Then, the user may customize the visualization by selecting the k scaling factor desired (i.e., # effects, # independent samples, or # articles) and the effect size scaling factor (i.e., raw or absolutized correlation). Note that, for some meta-science questions, the absolutized correlation may be more appropriate. For example, imagine I were interested in understanding how well, overall, attitudes/evaluations predict performance. Certainly, some of those attitudes are negatively valenced (e.g., perceived coworker incivility) and others are positively valenced (e.g., job satisfaction). In this case, the absolutized effect size would provide a partial solution (although also at risk of overestimation).
metaBUS

metaBUS.org | contact us | terms of service | privacy policy | Commercial use prohibited

Welcome to the metaBUS (beta) desktop interface!

Please start by viewing our video tutorial.

About metaBUS

  • What metaBUS is
    metaBUS is a platform for curating, locating, and rapidly summarizing research findings. Our processes leverage technologies to semi-automate the extraction of findings (e.g., correlation matrices) from documents and hand-coding to augment and error-check the extracted contents. At the core of the platform are several key enablements including (1) a database of approximately 1,000,000 effect sizes and associated metadata (as of December, 2016), (2) a revised taxonomy/ontology containing thousands of topics (e.g., job satisfaction; turnover intent...), (3) software to allow rapid, flexible queries (using text and/or taxonomy segments), and (4) R Statistics integration using several packages (e.g., metafor; Viechtbauer, 2010). In short, the metaBUS platform allows rapid search and exploration of a scientific space at the level of individual research findings. Currently, the platform encompasses portions of applied psychology and OB/HR management (with other relevant disciplines, such as social/personality psychology, included incidentally). The metaBUS team is currently expanding into related fields and welcomes discussion to expand further.
     
  • What metaBUS is not
    metaBUS is not an instant systematic review machine; rather, it is a search engine. Commonly used search engines regularly make errors of omission and commission. Furthermore, the contents of the metaBUS database are not double-coded. It is important that users of the platform are aware that the rapid summaries provided by metaBUS likely (1) lack completeness, (2) contain errors and, thus, (3) are not necessarily on par with results obtained from a state-of-the-science systematic review. However, with some error-checking, query results from metaBUS can certainly assist research teams in conducting systematic reviews (e.g., by facilitating the location of findings that were overlooked during literature search). The contents might also serve to provide first-pass estimates for relations of interest (or for benchmarking purposes) when a large collection of findings is returned. Using a variety of filters and other options, the platform provides output useful for “science of science” type research.
     
  • How is metaBUS funded?
    The metaBUS team is extremely grateful to its sponsors, without whom the project would not be possible:
    • Canadian Centre for Advanced Leadership in Business
    • Digging into Data Challenge (National Endowment for the Humanities)
    • National Science Foundation
    • SHRM Foundation
    • Social Sciences and Humanities Research Council
    • Virginia Commonwealth University Presidential Research Quest Fund
       
    • We are also extremely grateful for the efforts of 20 or so doctoral students and affiliates we’ve worked with over the years (see current contributors here)
    • Finally, this project would not have been possible without the "open" spirit backed by the R statistics community and many stackoverflow questions we floated along the way. metaBUS relies on the following R packages/libraries: compute.es, dplyr, DT, ggplot2, googleVis, gvlma, Hmisc, MAc, metafor, mongolite, plyr, RColorBrewer, readxl, rmongodb, rworldmap, shiny, shinydashboard, splines, and survival.
Understanding the results...