How to Determine the Number of Simulations in Sensitivity Analysis?

[Nub_User]

Hi colleagues,

I'm conducting a sensitivity analysis for a dynamic model with probabilistic variables as part of an academic research project and need to determine the necessary number of simulations.

By default, Vensim sets 200 simulations, and I used this value in my analysis. However, my reviewer questioned this choice, arguing that I had not explicitly justified it. I’d like to understand the key criteria for choosing a specific number.

• Are there established methods for determining the required number of simulations?

• What metrics can assess convergence to ensure sufficient simulations?

• Any academic references on this topic?

I’d appreciate any insights or recommendations.

Thank you!

[tomfid]

Good question.

There are probably some formal ways to do this, but I'd suggest an experimental approach: do a larger simulation (10x if possible) and test whether the marginal distributions of some key outcomes are the same. You can use the stats tool or the bar chart histogram to compare. To formalize, you could export the data and use something like the Kolmogorov-Smirnov test.