There is a study which claims that "learning by teaching" has a 90% retention rate. However, the study is no longer available and it is not clear what kind of research was actually done to support that claim. The legitimacy of that particular study is the subject of a related question on Skeptics.
Are there other studies that are available, proving or disproving that "learning by teaching" is more efficient than other learning methods?
According to one source, a man named Edgar Dale developed that retention rate pyramid. Edgar Dale had been both a teacher, and a school Superintendent.
The wikipedia article makes an interesting statement regarding this:
However, Dale included no numbers and did not base his cone on scientific research, and he also warned readers not to take the cone too seriously. The numbers may have originated as early as the 1940s, when a scholar at the University of Texas at Austin created visual aids for the military.
Wikipedia points to another article written by a Dr. Thalheimer, whose PhD was in Human Learning and Cognition, which is possibly a better resource. He makes this statement in his introduction:
People do NOT remember 10% of what they read, 20% of what they see, 30% of what they hear, etc. That information, and similar pronouncements are fraudulent. Moreover, general statements on the effectiveness of learning methods are not credible---learning results depend on too many variables to enable such precision. Unfortunately, this bogus information has been floating around our field for decades, crafted by many different authors and presented in many different configurations, including bastardizations of Dale's Cone.
Dr. Thalheimer goes further:
The bogus percentages were first published by an employee of Mobil Oil Company in 1967, writing in the magazine Film and Audio-Visual Communications. D. G. Treichler didn’t cite any research, but our field has unfortunately accepted his/her percentages ever since. NTL Institute still claims that they did the research that derived the numbers. See my response to NTL.
Michael Molenda, a professor at Indiana University, is currently working to track down the origination of the bogus numbers. His efforts have uncovered some evidence that the numbers may have been developed as early as the 1940's by Paul John Phillips who worked at University of Texas at Austin and who developed training classes for the petroleum industry.
That's interesting... does marketing voodoo usually stick around in the public mind longer than original research? Anyways, he re-emphasizes the importance of listing citations and the importance of understanding where research originally comes from, up to and including context. There were two guilty parties, according Dr. Thalheimer; of course, the original people who generated the tables, but also the folks who disseminated the research without checking the citations.
Thalheimer goes on to suggest that the information floating around in his industry can not be trusted.
It tells us that even our most reputable people and organizations may require the Wizard-of-Oz treatment---we may need to look behind the curtain to verify their claims.
So, how does one answer your question? Well, the US News rankings show Stanford as being the top University for psychology, at the moment. So, I'll take that at face value; because I don't have better information to go on, and assume that there are some decent researchers on cognition in Stanford. Out of the researchers who study cognition, which one's seems like they might be doing research on human learning? After I find one, I might be able to identify some of the journals in the field. They might even be some of the "top" journals in the field.
Part of the reason why this trick works, is that professors and fledgling professors usually keep a curriculum vitae, or CV. And often, listed on their CV is a set of references to papers that they have written in the past. Some are even so kind as to showcase some of the papers that they believe to be part of their best work to date. In some cases, this can act as a kind of roadmap to show you how to become a professor.
So this procedure pulls up a list of possible resources for journals that might have articles on that topic:
Cognition
Developmental Science
Developmental Psychology
Perspectives in Psychological Science
Here are two possible resources:
Frank, M. C., Tenenbaum, J. B., & Gibson, E. (2013). Learning and long-term retention of large-scale artificial languages. PLoS ONE, 8, e52500.
Shafto, P., Goodman, N., & Frank, M. C. (2012). Learning from others: The consequences of psychological reasoning for human learning. Perspectives in Psychological Science, 7, 341–351.
That might be enough of a seed to begin to answer your question. Let's begin by trying to read Shafto et al's paper in the journal Perspectives in Psychological Science. The official version is behind a paywall, but a version of the paper is freely available from Goodman's web page at Stanford.
In the abstract, Shafto et al says:
From early childhood, human beings learn not only from collections of facts about the world, but also in social contexts: from observation of other people, from communication, and from explicit teaching.
Beyond the abstract, in the 'Learning from others: The consequences of psychological reasoning for human learning', Shafto et al states:
Children are often compared to scientists, but even a perfect scientist, using experiments alone, would struggle to rediscover all of human knowledge in the span of one lifetime. How then are children able to acquire a good fraction of this knowledge in just a few years? The answer must be that children do not rediscover everything—they use their ability to reason intuitively about other people to learn what others already know.
Below figure 1, Shafto et al make the statement:
Research on human learning paints a very different picture of how datapoints are selected. A wide variety of approaches have pointed to people and their intentions, as an important factor in learning, highlighting that data are chosen rather than random (Bruner, 1966; Vygotsky, 1978) and that observed data are often the consequence of goal-directed actions (Bandura, Ross, & Ross, 1961; Gergely & Csibra, 2003; Meltzoff & Moore, 1977) or intentional communication/teaching (Coady, 1992; Csibra & Gergely, 2009; Harris, 2002; Tomasello, 1999; Tomasello, Carpenter, Call, Behne, & Moll, 2005).
So what can we gather from this? Within the field of psychology/cognition, there is discussion in the literature to support the possibility that people do, in fact, attempt to describe the process of learning by teaching. But the focus of Shafto et al's paper is not on trying to find the learning retention rates for different types of learning. In fact, nowhere in the paper does it even mention the term 'retention'. The focus of their paper, as described in the abstract, is rather to develop a formal framework for how a learner might observe someone's actions and infer from those how much that person knows about the world.
The focus of this paper could potentially be useful in trying to develop a learning model for an artificial intelligence.
In the conclusion, the authors make an interesting statement:
Our approach suggests that intuitive psychological reasoning potentially provides a very strong lever by which learners may capitalize on others’ knowledge to learn about the world (see also Coady, 1992; Csibra & Gergely, 2009; Harris, 2002; Tomasello et al., 2005). The key difference from previous formal approaches is how one views other people. If other people are viewed as random or even malicious in choosing their actions, then learning will likely be very difficult. However, if people are viewed as approximately rational, goal-directed agents, or as knowledgeable and helpful teachers, then the learning problem becomes much more tractable.
This statement appears to mean that the ability to learn by observing an observee's actions depends on the observer's assumed framework. If the observer assumes that the person whom they are observing is acting randomly, or perhaps that they are malicious in their actions; then it's more difficult to create a model to represent who they are observing. It is more difficult to learn. I'm not convinced that my interpretation of this statement is correct, given the context.
This paper also makes some references to studies done in the past. We could follow up by reading some of these as well. You may find this necessary, as this paper doesn't address your question directly.
