Defining Outcome Payoffs

In order to perform Outcome Optimization you will need to define a payoff (or payoffs for multicriteria optimization). To do so use the Payoff Specs which you open by first clicking on the Model Analysis Tools tab of the properties panel (with nothing selected in your model). Then click Payoff on the tabs that appear at the top.

.

A payoff is, in short, a score that is applied to a model run. The optimizer needs this to determine if one run is better than another. A run with a higher payoff is considered to be better than one with a lower payoff during optimization.

Payoffs are based on the value of model variables. One common definition is to use the value of a variable (such as cumulative profit) at the end of the simulation. Another is to look at the cumulative value of a variable (such as profit) over the course of the simulation. These two examples will give the same results, and the advantage of the second is that you don't need to add in additional model structure to make the computation.

Of course a payoff might have more than one component, and different components may contribute to it differently. Some components may be more important than others, and some may be bad things while others are good things. For example, suppose you want to include both the number of people living in poverty and average income in your payoff. Poverty is, presumably a bad things, while income would be good. In addition, the unit of measure for the number of people living in poverty is person, while average income would be something like $/year/person. To combine those into a single payoff you would need to give a negative weight to people living in poverty (say -1) and a positive weight to average income (say 1000). That way if there were 10 million people living in poverty and the average income was ten thousand then both would contribute equally to the payoff (which would be 0 at the baseline). Setting up weights so that each component of a payoff contributes equally is a simple rule of thumb for getting started. Those weights can then be adjusted up or down to reflect priorities.

Note that the actual value of the payoff does not really matter. Weights of -1 and 1000 would give the same results as weights of -0.1 and 100. The only time this will have an effect is if you are using an optimization type that lets you specify a tolerance or convergence criteria related to the payoff - in this case that value would need to be scaled to give the same results.

Multicriteria Payoffs

Since a payoff can have multiple components it is, in a sense, already multicriteria. However, the big distinction between single criteria and multicriteria optimization is that multicriteria optimization does not find a single best point Instead, it identifies a set of possibilities and allows stakeholder to exercise judgment in selecting between the possibilities. This selection, in which a variety of different performance metrics can be compared, is much easier that trying to agree on abstract weights that would go into a single payoff.