ABSTRACT
Researchers frequently conceptualize publication bias as a bias against publishing nonsignificant results. However, other factors beyond significance levels can contribute to publication bias. Some of these factors include study characteristics, such as the source of funding for the research project, whether the project was single center or multicenter, and prevailing theories at the time of publication. This article examines the relationship between publication bias and 2 study characteristics by breaking down 2 meta-analytic data sets into levels of the relevant study characteristic and assessing publication bias in each level with funnel plots, trim and fill (Duval & Tweedie, 2000a, 2000b), Egger's linear regression (Egger, Smith, Schneider, & Minder, 1997), cumulative meta-analysis (Borenstein, Hedges, Higgins, & Rothstein, 2009), and the Vevea and Hedges (1995) weight-function model. Using the Vevea and Hedges model, we conducted likelihood ratio tests to determine whether information was lost if only 1 pattern of selection was estimated. Results indicate that publication bias can differ over levels of study characteristics, and that developing a model to accommodate this relationship could be advantageous.
