Does SimDec replace the underlying model?

No. It analyzes model outputs or suitable data and depends on their quality and assumptions.

Is a scenario probability always equal?

No. The four equal probabilities in the worked example are an explicit toy assumption, not a general property.

Does it prove which real-world factor causes an outcome?

Not automatically. Causal interpretation needs an appropriate model or study design beyond visual grouping.

AI Glossary / Data engineering and architecture / Intermediate

Simulation decomposition

Definition

Simulation decomposition, often called SimDec, is an uncertainty and sensitivity analysis approach that explains an output distribution by separating it into scenarios associated with influential input conditions. The scenarios can be shown visually, helping analysts see which combinations of inputs produce different output ranges. This entry concerns the named SimDec method, rather than every technique that divides a simulation into smaller computational tasks.

What it means

A simulation may show a wide range of possible outcomes without explaining which input combinations contribute to that range. SimDec adds a structured view of those relationships. For example, it can help distinguish outcomes associated with low demand and high capacity from outcomes associated with high demand and low capacity. The interpretation remains conditional on the model, input distributions, and data; a compelling visualization does not establish that the model is correct.

How it works

The approach works with inputs and an output from simulations or suitable data. In the described implementation, sensitivity analysis helps identify influential inputs. Selected inputs are divided into states, such as low and high, and combinations of those states define scenarios. The output distribution is then decomposed by scenario to show how contributions overlap or separate. Analysts inspect the resulting patterns and check their robustness. Binning choices, sample coverage, and assumptions about the inputs influence what the visualization reveals.

Three practical examples

Capacity planning: an analyst explores how uncertain demand and processing capacity combine to produce queues in a model, then investigates the scenarios associated with long waits.
Engineering design: a team examines which combinations of uncertain material or environmental inputs contribute to undesirable simulated performance.
AI evaluation: a researcher explores modeled or observed performance variation across defined input conditions, while avoiding causal claims that the analysis design cannot support.

Think of it this way

Think of sorting a mixed bowl of colored beads to see which groups contribute to the overall pattern. The total remains the same, but its composition becomes easier to inspect. The analogy is limited because SimDec groups outcomes using input conditions and analytical choices. The colors are explanatory labels, not independent proof that changing one input in the real world will cause the predicted change.

Worked example

Use a tiny original capacity model with backlog = max(demand − capacity, 0). Demand can be 80 or 120 units, and capacity can be 90 or 110 units. The four combinations produce backlogs of 0, 0, 30, and 10 respectively. If each combination is equally likely, the mean backlog is (0 + 0 + 30 + 10) / 4 = 10. The mean alone hides that all positive backlog occurs in the high-demand scenarios, and lower capacity worsens that outcome. Grouping the results by demand and capacity makes this interaction visible. This four-point illustration explains scenario decomposition; it is not a full Monte Carlo analysis or a reproduction of SimDec's sensitivity-index calculation.

Illustrative case study

Fictional example: Hazel Logistics models a sorting operation with uncertain arrivals and processing times. An overall output histogram shows occasional long queues. The analysts use scenario decomposition to inspect which input combinations contribute to the upper range, then check whether those combinations are plausible and sufficiently sampled. They revise a questionable input assumption before considering operational changes. This invented case illustrates model interrogation, not a prediction for a real operation or a claim that the method guarantees an optimal staffing decision.

Common misconceptions and mistakes

Do not confuse the named method with parallelizing a simulation program. Do not treat colored regions as causal proof for empirical data. Input dependencies and model assumptions can affect interpretation. Avoid assuming that a visually distinct scenario is important without considering its frequency and consequence. A small sample may also make patterns unstable. Finally, a sensitivity analysis describes behavior under specified assumptions; it cannot validate omitted mechanisms or compensate for incorrect input data by itself.

Business use

SimDec can help communicate why a model produces a range of outcomes, especially when a single average or ranked importance list hides conditional behavior. Define the decision question, validate the model for that use, and make input assumptions visible. Compare the decomposition with other checks and test whether conclusions change under reasonable alternatives. For AI projects, use it as one analytical aid and keep the distinction between explanation of observed patterns and evidence about interventions explicit.

Check your understanding

A decomposition shows that high delays are concentrated in one scenario, but that scenario contains only two observations. Should the pattern be treated as settled?

Answer: No. Sample coverage and uncertainty need investigation. The display can suggest a question without establishing a robust conclusion.

Related terms

Monte Carlo simulation
Sensitivity analysis
Model validation
uncertainty analysis
scenario analysis
input distribution
interaction effect

Learn next

reproduce the four-case backlog example and change the probabilities of the input combinations. Compare the mean with the scenario contributions. Then explore a larger sample and document how input states and their dependence affect the interpretation.

Sources

Primary sources consulted: Roy and Kozlova, Simulation Decomposition in Python, https://joss.theoj.org/papers/10.21105/joss.06713.pdf — summary and method explanation read. Method project repository, https://github.com/Simulation-Decomposition/SimulationDecomposition.jl — search excerpt consulted. The arithmetic example and fictional case are original teaching illustrations, not reported study results. Full editorial review remains pending.

Previous: Simulation component metadata Next: Simulation environment metadata