Visual Sensitivity Analysis in Guesstimate

Trying to understand how variations in one variable impact another? You could manually adjust the first and observe changes in the second, but this takes time and doesn’t account for multivariable effects.

We at Guesstimate have been thinking about this issue and have recently implemented a solution we consider both intuitive and powerful.

Let’s explain with an example of a personal monthly budget. There are several uncertain categories of expenses, the sum is the total cost.

With Guesstimate you would first create this model, then click the Analysis view option, and finally click on the Total Monthly Expenses output node.

Above each other metric a small green scatterplot will appear, displaying the relationships between those metrics, on the x axis, and the Total Monthly Expenses metric (the selected metric), on the y axis.

A simple monthly expenses model. Copy it here.

If you look closely you can see each graph has a positive correlation. This demonstrates that each input is correlated with Total Monthly Expenses, which makes sense. In cases where one’s food budget is unusually high, one’s total expenses will often be as well.

There’s also a small r² value on each scatterplot. These represent the fit of linear regressions on the scatterplot data and can range from 0 to 1. Inputs with high r² explain much of output uncertainty. If you are trying to better understand and change a given output, its inputs with high r² values are often the ones to focus on.

In this case, Miscellaneous expenses have an r² value of 0.57, which is far higher than any of the other expenses. Let’s hover over this node to see an expanded analysis section.

A simple monthly expenses model, with hover on the Miscellaneous node on the right.

In the expanded section on the right, the positive correlation between Miscellaneous expenses and Total Monthly Expenses becomes more apparent. You can also get an idea of what the Total Monthly Expenses would be for all possible values of Miscellaneous expenses. Instead of needing to manually adjust that parameter to understand its impact, you can simply open this graph.

Conclusion

This is a simple example, but even here we were able to gain some useful insights. The Analysis view is more revealing for larger models, where relationships are more mysterious. We encourage you to look through all kinds of models through this view and make your own discoveries.