Generating hypotheses: A method for the madness

In research, the general goal is to plunge off into the unknown and wrest a little order out of the apparent chaos. Broadly, you can describe the phenomenon as taking a diverse set of existing observations, building a coherent model that can explain or generate these observations (I think of it as a 'reality engine'--an intellectual device that can give rise to the world as we see it), and then a series of tests that will distinguish the proposed model from other models (which may or may not be explicitly declared). The goal is to identify that which is distinctive about a given model and reinforce or contradict these features. In some ways, the toughest part is teasing an interesting and meaningful model out of the morass of observation. Following is an algorithm that was developed on the basis of conversations with Roy Parker

In brief, to create a model for a given body of observation, proceed as follows:

  1. Choose an observation that strikes you in some way. Perhaps it's the 'Big Question' in the field, perhaps a result others have swept under the rug, perhaps an aspect in which you have especial expertise
  2. Generate a preliminary model--a view of the world, system, machine that can account for the occurrence of the observation
  3. With the model in hand, you must now see how it survives testing by repeated confrontation with other relevant observations. Grab another observation that impinges on the model in some way. Does the model gracefully account for it? If so, swell with pride, gain confidence in the correctness of the model, and continue challenging it with existing observation. Ideally, your model will still be standing when you run out of observations.
    But what if the new observation does not sit well with the model? There are several ways to proceed:
    • You can ignore the fact. While this sounds childish and silly, in research it is often not inappropriate. Perhaps the observation (which may even go by the name of 'fact' in your field of study) was made in error. Perhaps the correct observation was made, but its INTERPRETATION is overstated, misleading, wrongheaded, etc. Perhaps it reflects a subtlety or complexity of the system that your simple model does not and need not address. There exist facts that we are simply not ready to understand from our current wisdom--setting these aside can allow us to progress. HOWever, as JFK may have said "Forgive your enemies--but never forget their names". Overlooked facts should be nagging your consciousness all the time. They are flaws in your model. And you CERTAINLY are only allowed to 'play the overlook card' a very few times!
    • You can modify your model. Perhaps it can be salvaged by tweaking an aspect. This is best. If you have to create a special 'exception mode' to accommodate the inconvenient observation, your model begins to become unwieldy. A good model explains the world in the simplest way; accumulated wisdom is that the world works on simple rules, and complicated rules generally mark a poor understanding. Witness the difficulty that Earth-cenric systems had in describing the orbits of the sun and planets around the Earth, vs. the simplicity of the 'Sun at the center' Copernican model.
    • You can reject your model. Return to Step 1, choosing either a different starting observation or seek a fresh way of explaining the observation.

     

For you visual folk, I've tried to graphically display these ideas as follows:

Back to Thinking page