Explanations that cannot be easily modified to account for anything
Hard to Vary Explanation
The Core Idea
David Deutsch argues that what distinguishes good explanations from bad ones is that good explanations are hard to vary while still accounting for what they purport to explain.
A bad explanation is easy to vary—you can change the details arbitrarily and it still “explains” the phenomenon. A good explanation is constrained: all its details play a functional role, so changing any part breaks the explanation.
Examples
Bad Explanation (Easy to Vary)
Ancient myth: The seasons occur because Persephone spends part of the year in the underworld.
This is easy to vary:
- Why not 3 months instead of 6? Still works.
- Why Persephone and not some other deity? Still works.
- Why the underworld and not a distant mountain? Still works.
The details are arbitrary. You can swap them freely and the “explanation” still accounts for seasons.
Good Explanation (Hard to Vary)
Modern astronomy: Seasons occur because Earth’s axis is tilted 23.5° relative to its orbital plane, causing different hemispheres to receive varying amounts of sunlight as Earth orbits the Sun.
This is hard to vary:
- Change the tilt angle substantially? No more pronounced seasons.
- Remove the tilt? No seasons at all.
- Change Earth’s orbit to circular with no tilt? Wrong prediction.
Every detail plays a role. The explanation is tightly constrained by what it explains.
Why This Matters
Against “Just So” Stories
Many explanations in psychology, evolutionary biology, and social science are easy to vary—they can account for almost any observation by tweaking details. Hard-to-vary criterion helps identify these.
Testability Connection
Hard-to-vary explanations tend to be testable because they make specific predictions. If details matter, changing them changes predictions, giving you ways to test.
Not the Same as Simplicity
A hard-to-vary explanation might be complex. What matters is whether the complexity is doing work. Simple explanations can be easy to vary (like myths), and complex ones can be hard to vary (like quantum mechanics).
Application to Research
When evaluating theories or building models:
- Can I change this detail without breaking the explanation?
- If I can freely vary parameters to fit any data, is this really explaining?
- What would make this explanation fail?
Tensions and Questions
- How do we apply this to probabilistic/statistical explanations?
- What about historical explanations where contingent details matter?
- Can we formalize “hard to vary” mathematically?
Connection to My Work
This framework shapes how I evaluate:
- Computational models: Are parameters doing work or just fitting data?
- Theoretical frameworks: Are they constrained or can they account for anything?
- My own explanations: Am I telling just-so stories or making risky claims?
Key Sources
- Deutsch, D. (2011). The Beginning of Infinity: Explanations That Transform the World
- Deutsch’s TED talk on “A new way to explain explanation”