Why Validating Your Model Matters: Ensuring Reliable Results

1 Why Validate Model Results?

Results that withstand validation are more convincing. After mathematizing a problem, building a model, solving and interpreting it, one must not forget to validate the model; this step often reveals whether the modeler has truly grasped mathematical modeling.

2 Validating Model Parameters

During model building and solving, we choose reasonable parameters—for example, an interest rate c within a range a%‑b% for budgeting. Since future conditions are uncertain, we care not only about results under the chosen parameters but also how results change when parameters vary. If adjusting a parameter (e.g., replacing c with another plausible value) yields little change, the result is “insensitive” and the conclusion remains usable. If the result changes dramatically, the result is “sensitive,” indicating the parameter is critical and must be estimated carefully for reliable outcomes.

Sensitivity analysis examines how parameter variations affect results and is a common validation method. Beyond parameters, we may have alternative processing choices—for instance, different normalization methods for data. By altering the processing method and checking whether results remain consistent, we can further validate the model. Consistent outcomes across methods suggest reliability; large discrepancies warrant deeper investigation.

3 Checking Results Against Common Sense or Domain Knowledge

Model outputs are used to expand knowledge or guide practice, but results that clearly contradict facts cannot be accepted. “Common sense” varies among individuals; some may treat certain knowledge as common sense, others as specialized expertise. Modelers lacking domain understanding may produce implausible results. Validation can involve consulting literature, researching information, or discussing with knowledgeable peers.

4 Using Validation to Extend Results

Beyond confirming reasonableness and robustness, validation can broaden results. Validation is not merely a checklist; it can involve alternative methods or facts that provide a new perspective on the outcome, effectively adding new content that supplements and expands the original model.

In summary:

Model results become more substantiated through validation.

The modeling process becomes more complete with validation.

The modeling thinking becomes more comprehensive with validation.