Meta-Analysis in the Era of Big Data: Power and Heterogeneity

• In Event: 2149 - Big Data and New Methods in Sociological Research

Abstract

Over the past few decades, meta-analysis has been increasingly used to summarize empirical evidence from a collection of independently-carried-out primary studies. Because of often dramatically increased statistical power, the approach has been a key methodological tool for obtaining credible empirical evidence in many cases in a wide range of scientific fields. It is high time that quantitative sociological researchers are introduced and trained in meta-analysis, which has accumulated a large literature over the decades. This review presents the literature of meta-analysis we consider most relevant to sociological research. For examples, this review focuses on the findings in primary studies in the form of regression coefficients rather than correlation coefficients because regression coefficients are typically presented in a sociological paper. The review includes sections on the rationale, preparation, derivation, and estimation of the fixed- and random-effects meta-analysis; two empirical examples with computer codes, one for the fixed-effects model and the other for the random-effects model; an introduction to methods that analyze the source of heterogeneity across studies; and a proposal that outlines steps through which sociological research will benefit from meta-analysis and raise the level of credibility of sociological findings via a knowledge-cumulative process in the era of big data.

Authors

• Guangyu Tong, Duke University

• Guang Guo, University of North Carolina