The causes, consequences and detection of publication bias in psychiatry.

OBJECTIVE

Publication bias threatens the validity of published research, although this topic has received little attention in psychiatry. The purpose of this article is to produce a systematic overview of the causes and consequences of publication bias and to summarize the available methods with which it is detected and corrected.

METHOD

Empirical evidence for the existence of publication bias is reviewed and the following methods are applied to an illustrative case example from psychiatry: funnel plot analysis; the 'file drawer method'; linear regression techniques; rank correlation; 'trim and fill'.

RESULTS

Small studies are particularly susceptible to publication and related bias. All methods to detect publication bias depend upon the availability of a number of individual studies with a range of sample sizes. Unfortunately, large numbers of studies of varying sample size are not always available in many areas of psychiatric research.

CONCLUSION

Where possible researchers should always test for the presence of publication bias. The problem of publication bias will not be solved by anything other than a prospective trials register.