My vote is for reason 2: people are citing things they haven’t read. Famous example is Likert’s 1932 paper which, to be fair, was really difficult to find and paywalled until relatively recently. But these days is quite easy to find and relatively easy to read when found. By the way: most academics will very happily send a paper to anyone who asks them. The current academic journal / paywall structure currently mostly exists to make money for the publishers by profiting from the donated labour of academics, and in some cases fees that academics have to pay. There are some exceptions: excellent open-access journal. And another thought: I’ve ended up citing something that was there when I cited it but is now unavailable at the original place and I haven’t been able to find it anywhere else. So there’s a fourth explanation: 4. Was available, isn’t now