The Uncertainty Machine
I’m continually shocked by how many genuinely useful tools exist that I’ve never heard of. This week’s discovery makes Monte Carlo analysis surprisingly approachable. Thanks to Duncan Swallow for pointing me to this gem.
As I wrote about here, Monte Carlo is a powerful tool for analyzing the interaction of the parameters that feed your analysis. Here’s how it would work when calculating the initial speed of a motorcycle that braked for 50 feet, went down and slid for 25 feet, and then hit a car.
In this example, three parameters have notable uncertainty: decel during braking, decel during the slide, and the impact speed. A conventional range calculation would combine all three low-end values to produce a minimum speed, then combine all three high-end values to produce a maximum speed. Here's the basic calc, without ranges.
But, how likely is it that all three inputs land at their minimum values at the same time, or all land at their maximums? Monte Carlo addresses that question by running thousands, or millions, of trials with values sampled from the allowable ranges (or assigned distributions, like normal). The result is a detailed picture of which outcomes are most likely.
And that’s where this week’s tool comes in: the NIST Uncertainty Machine. Instead of buying fancy software or tending to parentheses in Excel (which I’m actually a big fan of), you can run a Monte Carlo analysis in your browser and produce legit results in a few minutes.
After setting things up and clicking run, you’re provided with the distribution shape, mean, standard deviation, confidence intervals… the works. You can download the results and a configuration file that stores your setup. Perfect for future reference or to share with another expert. The file for the above example can be accessed here.
Just drop it into the Introduction section, and you’ll see how I set things up and all the related results. It feels like magic, but just it’s tried-and-true science.
Thanks for reading, keep exploring!
Lou Peck
P.S. I have the distinct honor interviewing Terry Day next week. If there's anything you'd like to hear him discuss, please let me know.