Are models properly idea-generating machines, in which you start from what you think is the case and use the model-building process to generate new insights? Or are models merely filing systems--ways of organizing your beliefs, and whenever you find that your model is leading you to a surprising conclusion that you find distasteful the proper response is to ignore the model, or to tweak it to make the distasteful conclusion go away?

Both can be effectively critiqued. Models-as-discovery-mechanisms suffer from the Polya-Robertson problem: It involves replacing what he calls "plausible reasoning", where models are there to assist thinking, with what he calls "demonstrative reasoning". in which the model itself becomes the object of analysis. The box that is the model is well described but, as Dennis Robertson warned,there is no reason to think that the box contains anything real. Models-as-filing-systems are often used like a drunk uses a lamp post: more for support than illumination.

In the real world, it is, of course, the case that models are both: both filing systems and discovery mechanisms. Coherent and productive thought is, as the late John Rawls used to say, always a process of reflective equilibrium--in which the trinity of assumptions, modes of reasoning, and conclusions are all three revised and adjusted under the requirement of coherence until a maximum level of comfort with all three is reached. The question is always one of balance.


モデルは、あなたの考察を出発点としたモデル構築の過程で新たな洞察を生み出すという用途で利用できる、アイディアを適切な形で生成する機械なのだろうか? それとも、モデルは単なるファイリングシステムであって、あなたの考えを整理する手法に過ぎず、モデルから忌むべき意外な結論が導かれたならば、適切な対応はモデルを無視するか、もしくはその忌むべき結論が無くなるようにモデルをいじることなのだろうか?



ちなみにクルーグマンはこのデロングのモデル論に特に反応していないが、サマーズは、「He is broadly right in my view that models function both as discovery tools and as ways of organizing and codifying thought.」と述べて、デロングの考察を概ね肯っている。

失礼します。失礼します。 2016/01/05 21:54 古典的には、少し進んでかな…


これ以上踏み込むと、なんとか警察が来そうなので、どうしようかな 、ようそう かな、りょうし ようかな、
可能 かな、ムリだなムリゲーでる。失礼します。

トラックバック - http://d.hatena.ne.jp/himaginary/20160105/two_faces_of_models