# 2.3.2: Limitations and challenges

For all its strengths, ABM comes with real limitations, and they fall into four broad kinds: validating a model against the world, scaling it up without running out of computing power, communicating its results clearly, and keeping it transparent enough to reproduce[^1]. Being able to name these kinds matters because a limitation you have anticipated can be planned around, whereas one that catches you late in a project can undo the work.

Each tab below unpacks one of the four kinds. Select each challenge to explore what it involves:

::::{tab-set}

:::{tab-item} Model validation
Validating ABM presents unique challenges that distinguish it from other modelling approaches:

**Data availability issues:**
- Real-world data at the individual agent level is often unavailable or incomplete
- Aggregate data may not capture the heterogeneity needed for agent-level validation
- Longitudinal data tracking individual behaviour over time is rare and expensive

**Validation complexity:**
- The stochastic nature of models makes exact replication challenging
- Multiple models may produce similar aggregate outcomes through different mechanisms
- Separating model validation from parameter calibration is complex
- Emergent behaviours are difficult to validate directly against empirical data
:::

:::{tab-item} Scalability and computational challenges
ABM simulations face significant computational constraints that limit their applicability:

**Computational intensity:**
- Large-scale simulations require significant computational resources
- Complex agent behaviours increase processing time exponentially
- Memory limitations restrict the number of agents and their state complexity
- Real-time applications may be impossible for complex models

**Scalability issues:**
- Adding more agents doesn't always scale linearly with computational cost
- Parallel processing can be challenging due to agent interdependencies
- Network effects can create computational bottlenecks
- Trade-offs between model realism and computational feasibility
:::

:::{tab-item} Communicating about your ABM
Effective communication of ABM results presents several challenges:

**Complexity communication:**
- Difficulty explaining emergent behaviours to non-technical audiences
- Challenge of communicating uncertainty and stochastic results
- Problem of oversimplification versus overwhelming detail
- Need to balance model accessibility with scientific rigour

**Stakeholder engagement:**
- Different stakeholders may focus on different aspects of the model
- Risk of stakeholders anthropomorphising agents inappropriately
- Difficulty conveying the limitations and assumptions of the model
- Challenge of managing expectations about model capabilities
:::

:::{tab-item} ABM transparency and reusability
Modern scientific practice demands transparency and reproducibility, which poses challenges for ABM:

**Documentation challenges:**
- Complex models require extensive documentation of assumptions and parameters
- Need to document agent behaviours, interaction rules, and environmental settings
- Requirement to explain model initialization and boundary conditions
- Challenge of maintaining documentation as models evolve

**Reproducibility issues:**
- Random number generation and stochastic processes must be carefully managed
- Software dependencies and version compatibility problems
- Platform-specific implementation differences
- Need for standardised reporting protocols (e.g., ODD protocol)

**Code sharing and reuse:**
- Models often contain proprietary or sensitive code
- Lack of standardisation in model structure and documentation
- Difficulty adapting models to new research questions
- Version control and collaborative development challenges
:::
::::

---

Now test whether you can recognise which kind each difficulty belongs to. In the activity below, match each difficulty to the limitation it illustrates, and treat the exercise as a way of learning to recognise trouble early rather than as a counsel of despair.

```{raw} html
:file: ../learning_activities_and_widgets/learning_activities/activity_2_3_2a_match_the_limitation.html
```

With ABM's strengths and limitations both in view, the next section pulls them together into a practical decision framework for judging when ABM is the right tool for a given problem.

## Key takeaways

```{admonition} Key takeaways
:class: important

- The flexibility of ABM carries real limitations, which fall into four kinds: validating a model against the world, scaling it within computational limits, communicating its results clearly, and keeping it transparent enough to reproduce.
- Because results can be sensitive to the assumed agent rules, a plausible-looking model may still be wrong, so validation cannot be taken for granted.
- Anticipating these limitations, and documenting the model accordingly, allows them to be planned around rather than discovered late in a project.
```

## References

[^1]: SEN1211 Course Materials, Lecture 2B, TU Delft. For enrolled students, detailed materials are available through the course platform.