Bài đọc 4
Imagine that twenty research teams independently test whether a particular vitamin improves memory. In reality it has no effect, but by pure chance one team finds what appears to be a positive result. That team publishes an exciting paper. The other nineteen, having found nothing, put their data away in a drawer. Anyone reading the scientific literature would conclude that the vitamin works.
This scenario illustrates what researchers call publication bias: the tendency for studies with positive or striking results to be published, while those that find no effect remain unseen. The psychologist Robert Rosenthal described it in 1979 as the "file drawer problem". It arises not from dishonesty but from incentives. Journals prefer findings that are new and surprising; researchers need publications to build their careers; and a result that says "we found nothing" rarely attracts attention.
The consequences became painfully clear during what is now called the replication crisis. In 2015, a large collaboration of researchers attempted to repeat one hundred studies published in leading psychology journals. Only around a third of the repeated studies produced statistically significant results similar to the originals. Similar concerns have since been raised in fields as varied as economics, cancer biology and medicine.
Publication bias matters well beyond academia. If doctors base treatment decisions on a literature that overstates the benefits of a drug, patients may receive treatments that are less effective than believed. For this reason, many medical authorities now require clinical trials to be registered publicly before they begin, so that trials that are never published can at least be counted.
Other reforms target the incentives themselves. In a format known as a registered report, scientists submit their research question and methods to a journal before collecting any data. If the plan is judged sound, the journal agrees in advance to publish the results, whatever they turn out to be. Some journals have also been created specifically to publish null results, and many researchers now share their data openly so that others can check their analyses.
None of this means that science is broken. The replication crisis was itself discovered by scientists using scientific methods, which is a sign of a system capable of correcting itself. But it does suggest that a single study, however impressive, should be treated as a first step rather than a final answer, and that the experiments which "fail" are often as informative as the ones that succeed.