Many of the references cited in scientific papers have not been read by the authors citing them, according to an analysis of how errors in citations propagate through the literature.

It isn't easy to establish directly — and truthfully — whether citations have been read. So Mikhail Simkin and Vwani Roychowdhury, two electrical engineers at the University of California, Los Angeles, decided to estimate author diligence by checking how often errors in citation lists are passed on through other papers. They reckon that scientists who have looked up and read a paper they want to cite are unlikely to misprint it in their own articles. The pair concluded that four out of five authors had not done their homework (M. V. Simkin and V. P. Roychowdhury Preprint cond-mat/0212043; 2002).

Simkin and Roychowdhury tracked all of the citations of a seminal 1973 paper on condensed-matter physics (J. M. Kosterlitz and D. J. Thouless J. Phys. C 6, 1181–1203; 1973). They looked at 4,300 citations of this paper and found that mis-citations of it were often identical to each other.

The probability of the same misprint occurring twice by chance is very small, so Simkin and Roychowdhury conclude that these errors are usually propagated by the reference being copied from someone else's citation list. The most common misprint appeared 78 times.

Based on the number of distinct misprints, the two researchers estimate that only 22–23% of citations followed from a reading of the original paper. And they postulate that this is typical of the scientific literature as a whole.

Others agree that lazy citation is a real issue. Sidney Redner, a physicist at Boston University who has studied citation statistics, says that he thinks the latest estimate of original reading may be on the low side. “I might have guessed at about 50%,” he says, but he agrees with the qualitative conclusion that many citations are unread.

How much does it matter? The Office of Research Integrity (ORI) at the US health department has recently drawn criticism for probing into this kind of practice, which is considered by some to be a kind of low-key misconduct (see Nature 420, 253; 200210.1038/420253a). If the problem is as widespread as the new study implies, it becomes harder to dismiss.

“I think it's very serious,” says Kay Fields of the ORI, although she sees it as a matter of negligence rather than misconduct. Fields points out that failure to consult the original source can also propagate misconceptions about what the source's conclusions actually are. “It's poor research practice, for sure,” she says.