Katie Hinde at ‘Mammals Suck’ writes
Only cite papers that you have read! DO NOT cite papers based on another publication’s report of them. Because every time that happens, a science fairy dies.
That’s an excellent principle. So why do I have a paper in press that cites a paper I haven’t read?
There are two reasons to cite a paper: as evidence for a claim, or to give credit to the authors for their research. In this case, I wanted to acknowledge that Takeuchi, in 1976, had come up with a modification of Akaike’s Information Criterion for possibly misspecified models. Our proposed criterion reduces to Takeuchi’s criterion under iid sampling.
Takeuchi wrote in Japanese. Not surprisingly, my campus library doesn’t stock the journal. I could probably get it via interlibrary loan, but to say I don’t read Japanese would be a massive understatement.
When you’re only citing a paper for credit there’s less risk. It’s still not ideal – eg, Lakatos pointed out in the second section of Proofs and Refutations that lots of people attribute the concept of uniform convergence to mathematicians who really just failed to notice it was an assumption – but it’s not going to lead to huge gaps in your arguments. Our arguments rely on fairly simple properties of quadratic forms, and we developed them ourselves. I’m not citing to shift the blame to Takeuchi if he was wrong, I’m crediting him for having (partly) got there first.
I feel reasonably safe with the reference, since I’ve seen it cited in two publications by Japanese researchers, and my original source in English was a book co-authored by Nils Lid Hjort, who’s careful about that sort of thing.
Still, I’m citing a paper I haven’t read.