Publication Bias

Sometimes investigators and research institutes spend a lot of time and efforts to conduct a study for examining a hypothesis, but it is possible that they reach to a non statistically significant conclusion, in this case, it is most likely to see unpublished the results of these studies. Studies show that the studies that reach a statistically significant conclusion are more likely to be published than those fail to reach significance and studies that reach significance conclusions are published more quickly than studies that do not reach significance. So why this is a case and this problem happen.

Study investigators may self-censor non-significant results, it takes time and energy to write up and publish study results, investigators may not wish to invest their time and energy in publishing studies that they feel are not exciting and instead put their efforts into more promising research. Rosenthal calls publication bias the file drawer problem as busy researchers may have file drawers full of results of no significant and unpublished studies.

In medicine most of the time pharmaceutical companies are sponsor for many clinical trials for evaluating the effects of a treatment, if they reach to non significant results, they are not interested to publish the results since they are the owner of the data and study. Journal editors and reviewer are less likely to accept the studies with non significant results. Publication bias affects the Meta analysis studies as well as any review of literature.

So what is wrong with publication bias?
The problem is that because of a fraction of studies on a subject that is available, it makes the field optimistic toward a treatment or a medicine.

Note: for writing this post, I used the slides of Meta Analysis course by Prof. Michael Lavalley