For example, when textbooks and courses discuss formal grammars, they typically focus on Noam Chomsky's generative grammars. Chomsky's model of grammar divorced structure from meaning, which he demonstrated with the nonsense sentence "Colorless green ideas sleep furiously." Since then, however, it has been observed by many commentators that a command of grammar as an independent from semantics is only really valid for people with training in grammar, and culturally not universal.
Take for example the early nineties computer game "First Samurai". Developed in the UK, they asked a Japanese translator what the game's title would be in the Japanese script, as they wanted this for the cover. The response was that it was impossible, because you can't rank samurai. Reportedly, they then asked how you would say the "first samurai" you see in the morning -- still impossible. In the end, they had to ask for another phrase to be translated and then take the symbol for "first" from that and place it next to the symbol for samurai.
There are dozens of similar anecdotes attested worldwide. In her 1978 book Children's Minds, Margaret Donaldson cites a report of an adventurer who asked a Native American to translate "the white man shot six bears today."
"'How can I do that?' said the Indian. 'No white man could shoot six bears in one day.'"Donaldson roundly rejects the idea that this grammaticality sense is anything more than a result of our education.
The other major blow to Chomsky's model was Lucien Tesnière's valency/dependency grammars. Whereas Chomsky built his trees based on part-of-speech only, Tesnière identified that certain words had to be accompanied by certain other features, and could optionally be qualified by additional ones.
Chomsky split his basic sentence trees into subject and predicate, as was the norm at the time. Tesnière instead argued for "verb centrality", putting the verb at the top of the tree. This meant that the verb in Chomsky's model had a direct link to the grammatical subject. It is trivially obvious that this is a superior model, because with no direct link, Chomsky's model essentially claims that "*He say I does" would pass a grammaticality test. Now I'm sure Chomsky at some point will have presented a round-about argument to say why that's not acceptable, quite simply Tesnière's model was better. Tesnière's model is widely accepted, and it's key to a lot of computer-based language techniques.
And yet when I studied language, lots of space was presented to Chomsky, and I have no recollection of seeing the name Tesnière or talk of valency or dependency grammars. When I briefly studied formal grammar in computing, lots of time was given to Chomsky, and dependency grammars were mentioned only in passing. To me, verb centrality was an obvious notion, and every time Chomskyan grammar was presented to me, I wanted to put the verb at the top. It wasn't until about four years ago that I picked up a book on Computational Linguistics/Natural Language Processing and was introduced to Tesnière's theories.
This is a very dangerous state of affairs -- students are being taught outdated, disproven theories instead of the current state of the art. In education, one of the best examples would be Piaget, whose theories have been proven wrong time and again, but are still one of the main focuses in most introductions to childhood development.
Why? The typical answer is that to understand the current system, we have to understand the underlying theories they're built on. This, I'm afraid, is not true. Or rather, it is true, but the underlying theories of modern grammar are derived from Chomsky, and the underlying theories of modern childhood development are derived from Piaget. By teaching the original theories, we end up holding back development in the field: most courses spend so long talking about the outdated theories that they don't leave time to fully discuss the current ones and the students leave the courses with a working model of the wrong theories. We therefore spend a lot of time debating the same thing as the generation before.
Certainly, we don't do this in the physical sciences. No-one would suggest that in order to learn about the big bang theory we first have to learn about the theories that predated it. Such theories are clearly of some interest, but are best restricted to specific modules on the history of science.
The problem in the social sciences seems to be a reluctance to rewrite part of a major theory based on subsequent observations and refinement. It appears that to be genuinely influential in social sciences, you cannot simply do incremental improvement, and instead must write a new theory practically from the ground up. In doing so, you are guaranteed posterity, because your grand theory will continue to be published, cited, repeated and taught as is long after all the elements it is built of are individually discredited -- the field will not allow anyone else to revise it for you.
Take Bloom's Taxonomy for instance. Even when it was first devised it was a bit of a kludge. The most common form seen today is still the simplest triangle form, and all versions and derivatives still hold the same ordering of "remembering" before "understanding" -- i.e. it preaches rote learning in line with the behaviorist thinking of Bloom's day, even though no modern school of thought actively professes a belief in rote learning as a useful mechanism.
The book Second Language Learning Theories by Mitchell, Myles and Marsden discusses the proliferation of theories in the introduction, and says "We believe that our understanding advances best when theories are freely debated and challenged among a community of scholars." I would certainly not dispute that, but I think we waste an awful lot of time when we discuss disproven theories and treat them as though they have equal merit to theories not yet disproven. We also, as I said earlier, do a lot of damage to the next generation of academics by preparing them to discuss the disproven theories rather than the current ones.
Worse, though, is the effect on non-academics. Teachers overexposed to outdated theories and not familiar with current ones are unable to take advantage of advances in the field and translate it into classroom practice.