Models without data are just entertaining animations. To answer research questions, you need systematic measurement and analysis.
Models Answer Questions with Evidence¶
Without data: “My segregation model seems to create clusters sometimes.” With data: “When tolerance is below 30%, segregation emerges in 95% of runs within 200 time steps.”
Without data: “Cooperation appears to work in this model.” With data: “Cooperation is stable when the population is smaller than 50 agents, but collapses above 75 agents.”
What Makes Good Model Data?¶
Systematic measurement:
- Track the same variables consistently
- Measure at regular intervals
- Run multiple times to account for randomness
- Test different parameter values systematically
Meaningful variables:
- Outcome measures: What you’re trying to explain (segregation level, cooperation rate, infection spread)
- Input parameters: What you’re varying (tolerance levels, payoff values, transmission rates)
- Process indicators: How things change over time (movement patterns, network formation, learning rates)