I wonder if the increase in the percentage of citation to older articles is related to the overall access situation getting better, though? Not that long ago, >10 yr old scientific articles were often quite difficult to access: shunted into "old journal" paper archives somewhere in a basement, from which they had to be specially requested. The library shelves typically only held 5-10 years of recent journal issues, and digitization efforts (at least in science) typically started with only the recent issues as well. To make matters worse, the old issues were rarely full-text indexed, so you might never even discover something relevant existed in the first place, much less try to access it. Nowadays an article from 1965 might randomly come up in a Google Scholar search, which was not the case in, say, 2002.
A more speculative hypothesis: this relates to the general "flattening" of the literature, as people relate to it more via search and less via subscriptions. People no longer really read journals as periodicals, or even use a specific journal's index as a search aid, but rather search entire archives for things. So it now matters less not only where something was published, but also when it was published.
Access is huge, and improved search algorithms probably help significantly too. Let's take the field of quantum cryptography as an example. How did it get started? This paper (the famous Bennett and Brassard, 1984 protocol, a.k.a. BB84):
This is quantum physics. Look at where it was published. Proceedings for a CPSC conference in Bangalore. Bangalore! This paper started an entire field that entire departments at multiple universities are devoted to now. If it hadn't been so radical at the time it would have made it into a high impact journal. Instead, IEEE conference proceedings, Bangalore.
I went through grad-school in this field. This paper started it all so, needless to say, I wanted to see an original copy of it. I never found one. These conference proceedings were not widely distributed. People passed around copies of copies of copies and there were, of course, reprints in other publications. I had to borrow a book that included this paper as a chapter from my supervisor, and that wasn't long ago. The only reason I was able to read this field-creating paper was because, even though the peer-reviewed journals that should have jumped at the chance to publish this work balked, other people recognized something good in it. Today, a pdf of the original is linked to by Wikipedia.
How much bold, radical, totally unpublishable (in any respectable journal) work is out there waiting to be dusted off and used to start new fields of study? BB84 could have slipped through the cracks. It's a safe bet that other such work actually did go unnoticed. We can only stand on the backs of those who came before us if we can access their work. I don't know if what we're experiencing today is an information revolution, but I'm willing to bet, in a few decades time, we'll be able to look back on the march of scientific progress and see a huge boost in research productivity when our ability to distribute and search through existing research started to catch up to the pace of its production.
> Various forms of functions described as Tanimoto similarity and Tanimoto distance occur in the literature and on the Internet. Most of these are synonyms for Jaccard similarity and Jaccard distance, but some are mathematically different. Many sources[3] cite an unavailable IBM Technical Report[4] as the seminal reference.
Huh! Worldcat says that a few copies are available. Now I want to see if I can get a copy. :)
I recently had to talk to a library archivist to track down what turns out to be a key paper which describes a practical topological encoding of a molecule for a computer. The author published it as a corporate white paper. If you look at the publications from the 1950s and 1960s you'll see people who cite the whitepaper, but as it wasn't part of the standard scientific literature, it wasn't saved, and no one has referenced that work for decades.
As it happens, he's also the person who coined "information retrieval", and was one of the influential founders of IR. Thus, his papers were archived, and still accessible.
One of the interesting things about hard-to-find papers like this is to see how the citations change over time, as typos get introduced and propagate.
"How much bold, radical, totally unpublishable (in any respectable journal) work is out there waiting to be dusted off and used to start new fields of study?"
No doubt quite a bit. I'm always reminded of the FFT, which was rediscovered in 1965 by Cooley and Tukey in the 1950s. It was first invented by Gauss, 160 years earlier, but not well known, "being published only posthumously and in neo-Latin".
As a culture, is it better to investigate more of the history, or spend time learning how to create this sort of thing?
Alternate hypothesis: less overall original research in general because we're not actually building on the new stuff, just continuously almost replicating results because no one bothered to read the new stuff.
A more speculative hypothesis: this relates to the general "flattening" of the literature, as people relate to it more via search and less via subscriptions. People no longer really read journals as periodicals, or even use a specific journal's index as a search aid, but rather search entire archives for things. So it now matters less not only where something was published, but also when it was published.