Metaphors and analogies are useful but may be dangerous if analogy is not homology. It is not clear in general where the line between analogy and homology falls and fails. Nonetheless, these metaphors and analogies seem to increase understanding in my undergraduate and graduate teaching, and so I offer them here.
The natural world is a great game with complex rules. The scientist can make limited observations and try to guess the rules, then make more observations to see if those rules are followed. The rules of checkers are far easier to guess from a few observations than are the rules of chess. Nature's rules are tougher yet. [I recall reading this metaphor in a column in Scientific American about two decades ago but have not been able to find it again. I would appreciate a citation if anyone knows the issue and page number in Scientific American.] I recently found this example again in Feynman (2002), a collection of wonderful essays, but I am not sure where this analogy originated.
The products of science are familiar in the form of textbooks that summarize theoretical and empirical knowledge and technologies that enable observations and other feats. Science as a factory is a useful metaphor because it emphasizes that seeing the product (e.g., a television set) does not give automatic understanding or appreciation of the production problems or methods. It is unfortunate that most introductory courses teach the products and not the process of science. Laboratory courses would seem an exception, but their exercises are usually so stereotyped and the outcomes so well known that they offer little insight into how new knowledge is obtained through comparison of competing but untested predictions against data.
Theory is constructed in the form of "if-then" statements about nature. If the world works like X, then we should observe Y. If, however, the world works like W, then we should see V. The goal is to come up with two or more believable "if-then" statements about the same process that predict measurably different outcomes.
Both are nonlinear, involve intense feedbacks, and may be nonstationary statistically over several ranges of time scales. It is easier for many ecology students to accept at the same time that past performance of the economy shows some statistical regularity but that mechanistic understanding of the economy is not sufficient for accurate prediction than it is for them to accept the same generalizations for ecosystems. While mechanistic understanding may still be lacking, both economies and ecosystems clearly warrant empirically based -- albeit imperfect -- management in the face of actual or potential anthropogenic perturbations. Conversely, this need for management is better accepted by non-ecologists for economies than it is for ecosystems, so the analogy can help in communications in both directions.
Feynman, R.P. 2002. The Pleasure of Finding Things Out and the Meaning of it All. Perseus Publishing, NY.