AN ESSENTIAL of science is that experiments should yield similar results if repeated. In recent years, however, some people have raised concerns that too many irreproducible results are being published (see chart 1). This phenomenon, it is suggested, may be a result of more studies having poor methodology, of more actual misconduct, or of both.
Or it may not exist at all, as Daniele Fanelli of the London School of Economics suggests in this week’s Proceedings of the National Academy of Sciences. First, although the number of erroneous papers retracted by journals has increased, so has the number of journals carrying retractions. Allowing for this, the number of retractions per journal has not gone up (chart 2). Second, as chart 3 shows, scientific-misconduct investigations by the Office of Research Integrity (ORI) in America are no more frequent than 20 years ago, nor are they more likely to find wrongdoing. Dr Fanelli’s point is not quite proved. The peak in reproducibility worries occurs after his retraction and ORI data run out. But it seems unlikely there would have been a sudden, recent shift in either.
As might be expected, countries with weaker misconduct policies than America’s—China and India, for example—are sources of more misdemeanours, such as the inappropriate copying or reuse of images like the gel patterns that result from DNA-sequencing experiments (chart 4). But, though the share of publications coming from these countries, and thus their contribution to the overall misdemeanour count, is increasing, there is, again, no evidence that the rate of bad behaviour there is rising.