The kind of oil you fry foods in has a big effect on how unhealthy it is. More >
Publication Bias May Hinder Research
There's a disturbing trend in the type of papers scholarly journals are publishing. A study from the University of Edinburgh has found a marked increase in the number of papers reporting positive correlations or links and a corresponding drop in the number that contained null or negative findings.
Negative findings – that a dietary supplement doesn't improve memory or that a lifestyle change doesn't increase longevity – are just as useful as positive ones, but fewer are getting published. This is thought to be because they attract fewer readers and are cited less often.
The study results are disturbing because they hint that journals are changing the nature of research. Some scientists have speculated that the practice of publishing only positive results will lead researchers to design only studies that have predictable outcomes and produce positive results through interpretation, selection or even manipulation of their data.
The Edinburgh study wasn't limited to scientific fields. It found the trend towards only publishing positive results was strongest in economics, business, clinical medicine, psychology, psychiatry, pharmacology and molecular biology. And the trend isn't limited to any one country; it's worldwide.
The researchers examined more than 4,600 papers published between 1990 and 2007. In that time span, the number of papers finding positive results increased from 70% to 86%, rising by nearly one-quarter.
Scientific studies are supposed to be tests of whether or not an unproven hypothesis is true. Their results aren't supposed to be foregone conclusions. A study finding that eating ice cream leads to weight gain isn't very enlightening. But that doesn't mean it won't get published.
The Edinburgh study results suggest that research is becoming less pioneering and that its objectivity is decreasing. In the study author's own words, "A system that disfavours negative results not only distorts the scientific literature directly, but might also discourage high-risk projects and pressure scientists to fabricate and falsify their data."
An article on the study was published in the September online issue of Scientometrics.
October 13, 2011