The inability in general to distinguish good outcomes from good decisions is also interesting, and has consequences for the creation of good decision processes. It can be illustrated with a simple example: Let us suppose that there is a historic data set of how often some children have played in the garden in each set of weather conditions. The combinations could be as follows:
Now, if we were unaware of today’s whether (e.g. due to travel away from home), but were told that today the children had played in the garden, what would this tell us about the whether at home?
The tree shown above can be “reversed” as shown:
Reversing the tree means that the branch order is reversed (i.e. Play in Garden is shown as the first branch), whilst the numerical fields are calculated to provide the same answer for any specific outcome. For example, by adding up in the first tree the figure 32% and 24%, one sees a 56% frequency of playing in the garden, which is shown as the first branch of the second tree (and so on).
In other words, to generalise this: Given sufficient historic data about joint occurrences of events, one can determine the frequency of one from the frequency of the other, or vice-versa. But to establish such a data set of course requires that the occurrence can be observed in an objective way.
Unfortunately, in the case of “good decisions” and “good outcomes”, it is can be difficult to observe or measure each and even harder to associate their joint occurrence. For example, there is often a long time frame in business contexts between an investment decision and the knowledge of whether it is profitable or not in practice; for some investments it could take 30 years or more to know, for example. The business, economic or political context could also change in ways that could not have been foreseen, and unusual risks may occur that also could not have been anticipated, even with robust risk assessment processes.