|
Abstract:
|
There is a substantial discussion regarding the consequences of publication bias. This has serious effects on when estimating e.g. treatment effects in meta studies. One of the most popular tools to de- tect publication bias is the use of a funnel plot which plots treatment effects against study precision. A nonsymmetric funnel plot implies some systematic difference between studies with lower and higher pre- cision, whereof one reason may be publication bias. One way to deal with publication bias is to "trim and fi ll" the data of the funnel plot (such that it becomes symmetric) and then analyze the resulting data with standard meta analysis tools. The "trim and fi ll" involves estimating the number of papers rejected which is nontrivial (see Duval and Tweedie, 2000, JASA). In this paper, we introduce a new way to estimate a treatment effect which deals with the publication bias problem. As the method is consistent (and not necessarily unbiased) a Monte Carlo simulation investigates the small sample performance. An empirical example shows the usefulneness of the proposed method.
|
