Model Space Search
Overview
Understanding = finding a model that fits. Instead of settling on first model that seems plausible:
- Generate multiple competing models
- Score each against fit criteria
- Select the model with best support
This prevents premature closure on a suboptimal model.
Goal
Generate multiple possible models/explanations for a phenomenon, then search for the one that best fits the evidence.
Steps
Step 1: Describe the Phenomenon
Clearly state what you’re trying to model. What patterns, behaviors, or outcomes need explaining?
Output: Phenomenon description
Step 2: List Observations
Document all relevant observations/data. These are what models must account for.
Output: Observation list
Step 3: Generate Null Model
What’s the simplest explanation? This is the baseline to beat.
Output: Null model
Step 4: Generate Causal Models
For each plausible cause, build a model. State cause, mechanism, predictions.
Output: Causal models
Step 5: Generate Mechanistic Models
What mechanisms could produce the observations? Build models around different mechanisms.
Output: Mechanistic models
Step 6: Generate Analogical Models
What similar phenomena have known models? Import and adapt.
Output: Analogical models
Step 7: Compile Model List
List all generated models. State each clearly with:
- Core claim
- Mechanism
- Key predictions
- Required assumptions
Output: Master model list
Step 8: Score Each Model
For each model, score on criteria:
- Explains observations
- No contradictions
- Predictive accuracy
- Simplicity
- Generalizability
- Coherence
- Mechanism clarity
Output: Scored models
Step 9: Compare Models
Create comparison table. Identify which model wins on which criteria. Look for dominant model (wins on most).
Output: Model comparison
Step 10: Select Best Model
Select model with highest score. Note:
- Confidence (how much better than alternatives?)
- What would change the selection?
- What further evidence would help?
Output: Selected model
Step 11: State Implications
If this model is correct:
- What else should be true?
- What should we expect to see?
- What actions does this suggest?
Output: Model implications
When to Use
- Trying to understand a phenomenon
- Multiple explanations seem plausible
- Need to choose between theories
- Building mental models
- Scientific/analytical inquiry
Verification
- Multiple models were generated (not just favorite)
- Null model was included as baseline
- Models are clearly stated with mechanisms
- Scoring was consistent across models
- Best model accounts for anomalies
- Implications and next steps identified