Graduate Student Seminar

April Albertine

Title

Some methods for addressing publication bias in meta-analysis

Abstract

In meta-analysis, the science of combining the results of many different primary studies in order to arrive at an overall conclusion about the question of interest, the researcher must make a serious effort to collect all relevant available studies (ranging from peer-reviewed articles to master's theses) for inclusion. However, the difficulty may still remain that some studies may be unpublished and unavailable to the researcher. This is the well-known "file drawer problem," in which studies with nonsignificant results languish in the file drawers of authors and journal editors, who are presumably less likely to submit or accept for publication studies that fail to reject the null hypothesis. This publication bias may lead to an overall conclusion of significance when in fact, including the unavailable studies might have yielded an overall nonsignificant result. In order to account for this bias, we build on two approaches outlined and developed by Iyengar and Greenhouse (1988). The first approach determines the number of nonsignificant studies that must exist in order for the meta-analysis to yield a just barely nonsignificant result; we provide a correction. The second approach incorporates a parameter for publication bias into the likelihood of the effect size, and then estimates the publication bias parameter and the effect size using maximum likelihood estimation. We simplify the procedure using a normal approximation. Finally, we present an application of both approaches.

Sai Popuri

Title

Abstract

In this seminar, I will introduce different types of spatial data and will briefly describe properties such as stationarity and isotropy. Assessing these properties is crucial prior to analyzing the data as many statistical methods assume that data are stationarity or/and isotropy. Using a few examples of point-referenced data, I will discuss ways to visualize spatial data and assess stationarity and isotropy using graphical exploratory methods. In this process, a few popular R packages useful in analyzing spatial data will be introduced.