CALCULUS FOR MODELLING 13

Model Limitations

Every model is an approximation. Understanding its limitations is part of mathematical modelling, not an optional addition.

The main idea

\[\boxed{\text{useful model}\neq\text{complete representation of reality}}\]

A model can be informative while still omitting important details.

1. Structural limitations

The chosen equations may exclude mechanisms that matter in the real system, such as age structure, spatial variation, delays, networks or behavioural change.

2. Parameter uncertainty

Parameters estimated from data are rarely known exactly. Different plausible parameter values can produce different trajectories.

3. Data limitations

Observations may contain measurement error, missing values, reporting delays, selection effects or changes in data collection.

4. Numerical limitations

Approximate numerical methods introduce discretisation error. Poor step sizes or algorithms can create behaviour that belongs to the computation rather than the mathematical model.

5. Deterministic limitations

An ODE model gives one trajectory for fixed parameters and initial conditions. It does not explicitly represent random variation between possible realisations.

6. Extrapolation

A model fitted or justified over one range of conditions may not remain valid far outside that range.

7. Identifiability

Sometimes different parameter combinations produce very similar outputs. Then the available data may not determine individual parameters reliably.

8. Validation

Model outputs should be compared with observations not used in construction when possible. Agreement supports usefulness within the tested setting but does not prove universal correctness.

9. Sensitivity analysis

Varying parameters and assumptions reveals which inputs strongly influence conclusions and which conclusions are robust.

10. Model comparison

More complicated is not automatically better. A useful model balances realism, interpretability, available data and the purpose of the analysis.

11. Communicating uncertainty

Model results should be presented together with assumptions, uncertainty and relevant limitations, especially when used for decisions.

12. Bridge to probability

Some limitations concern uncertainty itself. Probability provides mathematical tools for describing uncertain quantities and outcomes, leading naturally toward stochastic models.

\[\boxed{\text{deterministic modelling}\rightarrow\text{uncertainty}\rightarrow\text{probability}}\]

13. The central idea

A model should be judged by whether it is appropriate for its intended question, not by whether it reproduces every detail of reality.