In ancient Greece, sailors who survived shipwrecks had their portraits displayed in a temple on Samothrace as a testament to the power of Neptune. When Diagoras of Melos was told that this proved that the gods insert themselves into the lives of men, he answered, "but where are they painted that are drowned?"
Today, showing only the rescued sailors would be called publication bias, the tendency of scientists to report findings that support some point (Neptune rescues sailors) but to bury examples (drowned sailors) that undercut it. It has existed for years, most seriously in the failure to publish studies that cast doubt on the safety or efficacy of new drugs.
Now, guardians of scientific probity are fighting back. A handful of journals that publish only negative results are gaining traction, and new ones are on the drawing boards.
"You hear stories about negative studies getting stuck in a file drawer, but rigorous analyses also support the suspicion that journals are biased in favor of positive studies," says David Lehrer of the University of Helsinki, who is spearheading the new Journal of Spurious Correlations.
"Positive" means those showing that some intervention had an effect, that some gene is linked to a disease -- or, more broadly, that one thing is connected to another in a way that can't be explained by random chance. A 1999 analysis found that the percentage of positive studies in some fields routinely tops 90 percent. That is statistically implausible, suggesting that negative results are being deep-sixed. As a result, "what we read in the journals may bear only the slightest resemblance" to reality, concluded Lee Sigelman of George Washington University.
Example: In the 1990s, publication bias gave the impression of a link between oral contraceptives and cervical cancer. In fact, a 2000 analysis concluded, studies finding no link were seldom published, with the result that a survey of the literature led to "a spurious statistical connection."
Keeping a lid on negative results wastes time and money. In the 1980s, experiments claimed that an antibody called Rap-5 latches onto a cancer-related protein called Ras, exclusively. Scientists using Rap-5 then reported the presence of Ras in all sorts of human tumors, notes Scott Kern of Johns Hopkins University. That suggested that Ras is behind many cancers.
Oops. The antibody actually grabs other molecules, too. What scientists thought was Ras alone was a stew of compounds. In part because the glitch was published in obscure journals, researchers continued to use Rap-5 and reach erroneous conclusions, says Dr. Kern.
"If the negative results had been published earlier, scientists would have saved a lot of time and money," adds Bjorn Olsen of Harvard Medical School, a founding editor, with Christian Pfeffer, of the Journal of Negative Results in Biomedicine.
After a slow start in 2002, that journal is receiving more and better papers, says Dr. Olsen. One found that, contrary to other reports, the relative length of the bones of a woman's index finger and ring finger may not be related to her exposure to testosterone in utero. Another found that a molecule called PYY doesn't have a big influence on body weight; another, that variations in a gene that earlier studies had associated with obesity in mice and in American and Spanish women isn't linked to obesity in French men or women.
That may sound like the set-up for a joke, but studies that dispute connections between a gene and a disease are among the most important negative results in biomedicine. They undercut the simplistic idea that genes inevitably cause some condition, and show instead that how a gene acts depends on the so-called genetic background -- all of your DNA -- which affects how individual genes are activated and quieted. But you seldom see such negative results in top journals.
Hence, Dr. Olsen's journal, which is full of studies disputing reported links between gene variations and disease. The Sod1 gene and inherited forms of Lou Gehrig's disease? Probably not. MTHFR and the age at which Huntington disease strikes? Uh-uh. PINK-1 and late-onset Parkinson's disease? No.
Hopefully, each of these reports kept researchers, including those at drug companies, from wasting time looking for ways to repair the consequences of the supposed genetic association. But it isn't clear that any would have been published without the new journal.
Questionable correlations between a gene and cancer are the bread-and-butter of NOGO, the Journal of Negative Observations in Genetic Oncology, which Dr. Kern edits. "Fully half (of discoveries) of novel mutations in tumors, we found, were not confirmed in the subsequent literature," he says. "You expect to see follow-ups if the claims held up, so the fact that we didn't casts doubt on the original claim. But that wasn't explicitly reported."
Why are scientists coy about publishing negative data? In some cases, says Dr. Kern, withholding them keeps rivals doing studies that rest on an erroneous premise, thus clearing the field for the team that knows that, say, gene A doesn't really cause disease B.
Which goes to show that in scientific journals, no less than in supermarket tabloids, you can't believe everything you read -- or shouldn't.