Types of scientific explanations

1. old-school law-based explanations. Names: Carl Hempel, Paul Oppenheim The rough idea is: explain an event by showing it necessarily follows from laws plus initial conditions. Carl Hempel’s Deductive-Nomological model If you can derive the event from the law-governed setup, you have explained it. Here’s an example event to explain: Why did my cup of water freeze? Well, the Law says:  Water freezes below 0°C. And the Initial Conditions were:

• The temperature on my porch is -10°C
• I left a cup of water on the porch.

Since we can formally deduce the event from this setup, congratulations, we have explained it.

Of course then you can ask “why does water freeze below 0°C?” and chain the laws deeper and deeper until you reach the most foundational laws of reality. Much of the history of physics is basically this, bottoming out in foundational principles like relativity, quantum mechanics, and the initial conditions of the Big Bang, from which every other observation is thought to be possible to derive.

Why isn’t this enough? Because the deductive-nomological explanation doesn’t account for causality.

Suppose you’re concerned that your son might be short. You come to your pediatric doctor for a consultation.

You: Why is my son only 1 meter tall?

Doctor: Well, the optics law is “When the distant source of light is at a 45 degree angle to the object, the shadow is as long as the object.” And the conditions are: the sun is at a 45 degree angle, and your son’s shadow is 1 meter long. From that we can formally deduce that your son must be 1 meter tall. That’s the only way it can be according to the laws of physics. Here, I explained it. Why are you angry?

Our dissatisfaction with explanations built purely on logical deduction is known in philosophy of science as a “flagpole problem”, and much of the 20th century history was spent looking for more satisfying explanations.

2. causal explanations. Names: Michael Scriven, Wesley Salmon, James Woodward The rough idea is that to explain an event is to show what brought it about. Salmon’s later causal-mechanical model says an explanation traces the causal processes and interactions that place the event into a “causal nexus.” (???) Woodward’s interventionism: C explains E if changing C by a suitable intervention would change E. Modern causal inference people get excited by interventions rather than bare correlations. Basically these models try to infer the direction of the causal arrows in their models by running side-to-side experiments with control groups. Good enough for labs, maybe not good for economics, history, or astronomy. But:

You: Why is my son only 1 meter tall?

Doctor: Our experiments indicate that limb stretching surgery increases height by as much as 20cm. Have you done any limb stretching on your son?

You: What? No!

Doctor: Well there’s your explanation. Why are you angry?

3. mechanistic explanations. This is the dominant style in a lot of wet lab biology. Names: New mechanists such as Peter Machamer, Lindley Darden, Carl Craver, and William Bechtel Mechanistic explanation is: showing how a “black box” phenomenon is decomposed into a set of parts linked by activities. In practice, it looks like a diagram where blobs are connected with arrows.
4. unificationist explanations. Main names: Michael Friedman and Philip Kitcher. Here the thought is not primarily “what caused this?” but “how many seemingly separate facts collapse into one pattern?” Newton explaining both apples and planets with one gravitational framework is the stock example. Maxwell unifying electricity and magnetism is another. Unificationists think explanation is good when it reduces the number of separate brute facts you must accept. Problem: an explanation can unify elegantly yet still fail to capture causal direction or manipulable structure.
5. pragmatic or contextual explanations Main name: Bas van Fraassen ??? what it does that others dont? where does it fail?
6. constraint-based explanations Names: Deutsch’s constructor theory, Marletto, any others? Constraint explanation is cast in terms of possible and impossible transformations: not just trajectories generated by laws, but which tasks can be caused to happen and which cannot. Conservation laws, no-cloning, impossibility of perpetual motion, lightspeed limit, thermodynamic restrictions: these are often more informative than a simple chronicle of what happened. Here, best explanations are those that narrow down your expected observations the most: “anything can happen” is a much worse explanation than “only this one specific thing can happen”.

In many fields, it can take a form of a dynamical-system explanation. It sounds like “this is the stable attractor state, all others are unstable”.